init

.gitattributes

@@ -1,35 +1,35 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
+# *.7z filter=lfs diff=lfs merge=lfs -text
+# *.arrow filter=lfs diff=lfs merge=lfs -text
+# *.bin filter=lfs diff=lfs merge=lfs -text
+# *.bz2 filter=lfs diff=lfs merge=lfs -text
+# *.ckpt filter=lfs diff=lfs merge=lfs -text
+# *.ftz filter=lfs diff=lfs merge=lfs -text
+# *.gz filter=lfs diff=lfs merge=lfs -text
+# *.h5 filter=lfs diff=lfs merge=lfs -text
+# *.joblib filter=lfs diff=lfs merge=lfs -text
+# *.lfs.* filter=lfs diff=lfs merge=lfs -text
+# *.mlmodel filter=lfs diff=lfs merge=lfs -text
+# *.model filter=lfs diff=lfs merge=lfs -text
+# *.msgpack filter=lfs diff=lfs merge=lfs -text
+# *.npy filter=lfs diff=lfs merge=lfs -text
+# *.npz filter=lfs diff=lfs merge=lfs -text
+# *.onnx filter=lfs diff=lfs merge=lfs -text
+# *.ot filter=lfs diff=lfs merge=lfs -text
+# *.parquet filter=lfs diff=lfs merge=lfs -text
+# *.pb filter=lfs diff=lfs merge=lfs -text
+# *.pickle filter=lfs diff=lfs merge=lfs -text
+# *.pkl filter=lfs diff=lfs merge=lfs -text
+# *.pt filter=lfs diff=lfs merge=lfs -text
+# *.pth filter=lfs diff=lfs merge=lfs -text
+# *.rar filter=lfs diff=lfs merge=lfs -text
+# *.safetensors filter=lfs diff=lfs merge=lfs -text
+# saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+# *.tar.* filter=lfs diff=lfs merge=lfs -text
+# *.tar filter=lfs diff=lfs merge=lfs -text
+# *.tflite filter=lfs diff=lfs merge=lfs -text
+# *.tgz filter=lfs diff=lfs merge=lfs -text
+# *.wasm filter=lfs diff=lfs merge=lfs -text
+# *.xz filter=lfs diff=lfs merge=lfs -text
+# *.zip filter=lfs diff=lfs merge=lfs -text
+# *.zst filter=lfs diff=lfs merge=lfs -text
+# *tfevents* filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,2 @@
+*.caffemodel
+*.pth

app.py

@@ -0,0 +1,92 @@
+import cv2
+import matplotlib.pyplot as plt
+import copy
+import numpy as np
+import gradio as gr
+from src import model
+from src import util
+from src.body import Body
+from src.hand import Hand
+
+
+def pose_estimation(test_image):
+    bgr_image_path = './test.png'
+    with open(bgr_image_path, 'wb') as bgr_file:
+        bgr_file.write(test_image)
+    # 加载估计模型
+    body_estimation = Body('model/body_pose_model.pth')
+    hand_estimation = Hand('model/hand_pose_model.pth')
+
+    test_image = bgr_image_path
+    oriImg = cv2.imread(test_image)  # B,G,R order
+
+    # oriImg = test_image
+
+    # 姿态估计
+    candidate, subset = body_estimation(oriImg)
+    canvas = copy.deepcopy(oriImg)
+    # 绘制身体姿态
+    canvas = util.draw_bodypose(canvas, candidate, subset)
+    # print(candidate)
+    # print(subset)
+    # detect hand
+    hands_list = util.handDetect(candidate, subset, oriImg)
+
+    all_hand_peaks = []
+    for x, y, w, is_left in hands_list:
+        # cv2.rectangle(canvas, (x, y), (x+w, y+w), (0, 255, 0), 2, lineType=cv2.LINE_AA)
+        # cv2.putText(canvas, 'left' if is_left else 'right', (x, y), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
+
+        # if is_left:
+            # plt.imshow(oriImg[y:y+w, x:x+w, :][:, :, [2, 1, 0]])
+            # plt.show()
+        peaks = hand_estimation(oriImg[y:y+w, x:x+w, :])
+        peaks[:, 0] = np.where(peaks[:, 0]==0, peaks[:, 0], peaks[:, 0]+x)
+        peaks[:, 1] = np.where(peaks[:, 1]==0, peaks[:, 1], peaks[:, 1]+y)
+        # else:
+        #     peaks = hand_estimation(cv2.flip(oriImg[y:y+w, x:x+w, :], 1))
+        #     peaks[:, 0] = np.where(peaks[:, 0]==0, peaks[:, 0], w-peaks[:, 0]-1+x)
+        #     peaks[:, 1] = np.where(peaks[:, 1]==0, peaks[:, 1], peaks[:, 1]+y)
+        #     print(peaks)
+        all_hand_peaks.append(peaks)
+
+    canvas = util.draw_handpose(canvas, all_hand_peaks)
+
+    plt.imshow(canvas[:, :, [2, 1, 0]])
+    plt.axis('off')
+    plt.savefig('./out.jpg')
+    # plt.show()
+    return './out.jpg'
+
+# Convert the image path to bytes for Gradio to display
+def convert_image_to_bytes(image_path):
+    with open(image_path, "rb") as image_file:
+        return image_file.read()
+
+# Gradio interface
+with gr.Blocks() as demo:
+    gr.Markdown("# Pose Estimation")
+    with gr.Row():
+        image = gr.File(label="Upload Image", type="binary")
+        output_image = gr.Image(label="Estimation Result")
+    submit_button = gr.Button("Start Estimation")
+    
+    # Run pose estimation and display results when the button is clicked
+    submit_button.click(
+        pose_estimation,
+        inputs=[image],
+        outputs=[output_image]
+    )
+    
+    # Clear the results
+    clear_button = gr.Button("Clear")
+    def clear_outputs():
+        output_image.clear()
+    clear_button.click(
+        clear_outputs,
+        inputs=[],
+        outputs=[output_image]
+    )
+
+if __name__ == "__main__":
+    demo.launch(debug=True)

model/body_pose_deploy.prototxt

@@ -0,0 +1,2976 @@
+input: "image"
+input_dim: 1
+input_dim: 3
+input_dim: 1 # This value will be defined at runtime
+input_dim: 1 # This value will be defined at runtime
+layer {
+  name: "conv1_1"
+  type: "Convolution"
+  bottom: "image"
+  top: "conv1_1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 64
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu1_1"
+  type: "ReLU"
+  bottom: "conv1_1"
+  top: "conv1_1"
+}
+layer {
+  name: "conv1_2"
+  type: "Convolution"
+  bottom: "conv1_1"
+  top: "conv1_2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 64
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu1_2"
+  type: "ReLU"
+  bottom: "conv1_2"
+  top: "conv1_2"
+}
+layer {
+  name: "pool1_stage1"
+  type: "Pooling"
+  bottom: "conv1_2"
+  top: "pool1_stage1"
+  pooling_param {
+    pool: MAX
+    kernel_size: 2
+    stride: 2
+  }
+}
+layer {
+  name: "conv2_1"
+  type: "Convolution"
+  bottom: "pool1_stage1"
+  top: "conv2_1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu2_1"
+  type: "ReLU"
+  bottom: "conv2_1"
+  top: "conv2_1"
+}
+layer {
+  name: "conv2_2"
+  type: "Convolution"
+  bottom: "conv2_1"
+  top: "conv2_2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu2_2"
+  type: "ReLU"
+  bottom: "conv2_2"
+  top: "conv2_2"
+}
+layer {
+  name: "pool2_stage1"
+  type: "Pooling"
+  bottom: "conv2_2"
+  top: "pool2_stage1"
+  pooling_param {
+    pool: MAX
+    kernel_size: 2
+    stride: 2
+  }
+}
+layer {
+  name: "conv3_1"
+  type: "Convolution"
+  bottom: "pool2_stage1"
+  top: "conv3_1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu3_1"
+  type: "ReLU"
+  bottom: "conv3_1"
+  top: "conv3_1"
+}
+layer {
+  name: "conv3_2"
+  type: "Convolution"
+  bottom: "conv3_1"
+  top: "conv3_2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu3_2"
+  type: "ReLU"
+  bottom: "conv3_2"
+  top: "conv3_2"
+}
+layer {
+  name: "conv3_3"
+  type: "Convolution"
+  bottom: "conv3_2"
+  top: "conv3_3"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu3_3"
+  type: "ReLU"
+  bottom: "conv3_3"
+  top: "conv3_3"
+}
+layer {
+  name: "conv3_4"
+  type: "Convolution"
+  bottom: "conv3_3"
+  top: "conv3_4"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu3_4"
+  type: "ReLU"
+  bottom: "conv3_4"
+  top: "conv3_4"
+}
+layer {
+  name: "pool3_stage1"
+  type: "Pooling"
+  bottom: "conv3_4"
+  top: "pool3_stage1"
+  pooling_param {
+    pool: MAX
+    kernel_size: 2
+    stride: 2
+  }
+}
+layer {
+  name: "conv4_1"
+  type: "Convolution"
+  bottom: "pool3_stage1"
+  top: "conv4_1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu4_1"
+  type: "ReLU"
+  bottom: "conv4_1"
+  top: "conv4_1"
+}
+layer {
+  name: "conv4_2"
+  type: "Convolution"
+  bottom: "conv4_1"
+  top: "conv4_2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu4_2"
+  type: "ReLU"
+  bottom: "conv4_2"
+  top: "conv4_2"
+}
+layer {
+  name: "conv4_3_CPM"
+  type: "Convolution"
+  bottom: "conv4_2"
+  top: "conv4_3_CPM"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu4_3_CPM"
+  type: "ReLU"
+  bottom: "conv4_3_CPM"
+  top: "conv4_3_CPM"
+}
+layer {
+  name: "conv4_4_CPM"
+  type: "Convolution"
+  bottom: "conv4_3_CPM"
+  top: "conv4_4_CPM"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu4_4_CPM"
+  type: "ReLU"
+  bottom: "conv4_4_CPM"
+  top: "conv4_4_CPM"
+}
+layer {
+  name: "conv5_1_CPM_L1"
+  type: "Convolution"
+  bottom: "conv4_4_CPM"
+  top: "conv5_1_CPM_L1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu5_1_CPM_L1"
+  type: "ReLU"
+  bottom: "conv5_1_CPM_L1"
+  top: "conv5_1_CPM_L1"
+}
+layer {
+  name: "conv5_1_CPM_L2"
+  type: "Convolution"
+  bottom: "conv4_4_CPM"
+  top: "conv5_1_CPM_L2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu5_1_CPM_L2"
+  type: "ReLU"
+  bottom: "conv5_1_CPM_L2"
+  top: "conv5_1_CPM_L2"
+}
+layer {
+  name: "conv5_2_CPM_L1"
+  type: "Convolution"
+  bottom: "conv5_1_CPM_L1"
+  top: "conv5_2_CPM_L1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu5_2_CPM_L1"
+  type: "ReLU"
+  bottom: "conv5_2_CPM_L1"
+  top: "conv5_2_CPM_L1"
+}
+layer {
+  name: "conv5_2_CPM_L2"
+  type: "Convolution"
+  bottom: "conv5_1_CPM_L2"
+  top: "conv5_2_CPM_L2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu5_2_CPM_L2"
+  type: "ReLU"
+  bottom: "conv5_2_CPM_L2"
+  top: "conv5_2_CPM_L2"
+}
+layer {
+  name: "conv5_3_CPM_L1"
+  type: "Convolution"
+  bottom: "conv5_2_CPM_L1"
+  top: "conv5_3_CPM_L1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu5_3_CPM_L1"
+  type: "ReLU"
+  bottom: "conv5_3_CPM_L1"
+  top: "conv5_3_CPM_L1"
+}
+layer {
+  name: "conv5_3_CPM_L2"
+  type: "Convolution"
+  bottom: "conv5_2_CPM_L2"
+  top: "conv5_3_CPM_L2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu5_3_CPM_L2"
+  type: "ReLU"
+  bottom: "conv5_3_CPM_L2"
+  top: "conv5_3_CPM_L2"
+}
+layer {
+  name: "conv5_4_CPM_L1"
+  type: "Convolution"
+  bottom: "conv5_3_CPM_L1"
+  top: "conv5_4_CPM_L1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu5_4_CPM_L1"
+  type: "ReLU"
+  bottom: "conv5_4_CPM_L1"
+  top: "conv5_4_CPM_L1"
+}
+layer {
+  name: "conv5_4_CPM_L2"
+  type: "Convolution"
+  bottom: "conv5_3_CPM_L2"
+  top: "conv5_4_CPM_L2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "relu5_4_CPM_L2"
+  type: "ReLU"
+  bottom: "conv5_4_CPM_L2"
+  top: "conv5_4_CPM_L2"
+}
+layer {
+  name: "conv5_5_CPM_L1"
+  type: "Convolution"
+  bottom: "conv5_4_CPM_L1"
+  top: "conv5_5_CPM_L1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 38
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "conv5_5_CPM_L2"
+  type: "Convolution"
+  bottom: "conv5_4_CPM_L2"
+  top: "conv5_5_CPM_L2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 19
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "concat_stage2"
+  type: "Concat"
+  bottom: "conv5_5_CPM_L1"
+  bottom: "conv5_5_CPM_L2"
+  bottom: "conv4_4_CPM"
+  top: "concat_stage2"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage2_L1"
+  type: "Convolution"
+  bottom: "concat_stage2"
+  top: "Mconv1_stage2_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage2_L1"
+  type: "ReLU"
+  bottom: "Mconv1_stage2_L1"
+  top: "Mconv1_stage2_L1"
+}
+layer {
+  name: "Mconv1_stage2_L2"
+  type: "Convolution"
+  bottom: "concat_stage2"
+  top: "Mconv1_stage2_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage2_L2"
+  type: "ReLU"
+  bottom: "Mconv1_stage2_L2"
+  top: "Mconv1_stage2_L2"
+}
+layer {
+  name: "Mconv2_stage2_L1"
+  type: "Convolution"
+  bottom: "Mconv1_stage2_L1"
+  top: "Mconv2_stage2_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage2_L1"
+  type: "ReLU"
+  bottom: "Mconv2_stage2_L1"
+  top: "Mconv2_stage2_L1"
+}
+layer {
+  name: "Mconv2_stage2_L2"
+  type: "Convolution"
+  bottom: "Mconv1_stage2_L2"
+  top: "Mconv2_stage2_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage2_L2"
+  type: "ReLU"
+  bottom: "Mconv2_stage2_L2"
+  top: "Mconv2_stage2_L2"
+}
+layer {
+  name: "Mconv3_stage2_L1"
+  type: "Convolution"
+  bottom: "Mconv2_stage2_L1"
+  top: "Mconv3_stage2_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage2_L1"
+  type: "ReLU"
+  bottom: "Mconv3_stage2_L1"
+  top: "Mconv3_stage2_L1"
+}
+layer {
+  name: "Mconv3_stage2_L2"
+  type: "Convolution"
+  bottom: "Mconv2_stage2_L2"
+  top: "Mconv3_stage2_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage2_L2"
+  type: "ReLU"
+  bottom: "Mconv3_stage2_L2"
+  top: "Mconv3_stage2_L2"
+}
+layer {
+  name: "Mconv4_stage2_L1"
+  type: "Convolution"
+  bottom: "Mconv3_stage2_L1"
+  top: "Mconv4_stage2_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage2_L1"
+  type: "ReLU"
+  bottom: "Mconv4_stage2_L1"
+  top: "Mconv4_stage2_L1"
+}
+layer {
+  name: "Mconv4_stage2_L2"
+  type: "Convolution"
+  bottom: "Mconv3_stage2_L2"
+  top: "Mconv4_stage2_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage2_L2"
+  type: "ReLU"
+  bottom: "Mconv4_stage2_L2"
+  top: "Mconv4_stage2_L2"
+}
+layer {
+  name: "Mconv5_stage2_L1"
+  type: "Convolution"
+  bottom: "Mconv4_stage2_L1"
+  top: "Mconv5_stage2_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage2_L1"
+  type: "ReLU"
+  bottom: "Mconv5_stage2_L1"
+  top: "Mconv5_stage2_L1"
+}
+layer {
+  name: "Mconv5_stage2_L2"
+  type: "Convolution"
+  bottom: "Mconv4_stage2_L2"
+  top: "Mconv5_stage2_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage2_L2"
+  type: "ReLU"
+  bottom: "Mconv5_stage2_L2"
+  top: "Mconv5_stage2_L2"
+}
+layer {
+  name: "Mconv6_stage2_L1"
+  type: "Convolution"
+  bottom: "Mconv5_stage2_L1"
+  top: "Mconv6_stage2_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage2_L1"
+  type: "ReLU"
+  bottom: "Mconv6_stage2_L1"
+  top: "Mconv6_stage2_L1"
+}
+layer {
+  name: "Mconv6_stage2_L2"
+  type: "Convolution"
+  bottom: "Mconv5_stage2_L2"
+  top: "Mconv6_stage2_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage2_L2"
+  type: "ReLU"
+  bottom: "Mconv6_stage2_L2"
+  top: "Mconv6_stage2_L2"
+}
+layer {
+  name: "Mconv7_stage2_L1"
+  type: "Convolution"
+  bottom: "Mconv6_stage2_L1"
+  top: "Mconv7_stage2_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 38
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mconv7_stage2_L2"
+  type: "Convolution"
+  bottom: "Mconv6_stage2_L2"
+  top: "Mconv7_stage2_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 19
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "concat_stage3"
+  type: "Concat"
+  bottom: "Mconv7_stage2_L1"
+  bottom: "Mconv7_stage2_L2"
+  bottom: "conv4_4_CPM"
+  top: "concat_stage3"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage3_L1"
+  type: "Convolution"
+  bottom: "concat_stage3"
+  top: "Mconv1_stage3_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage3_L1"
+  type: "ReLU"
+  bottom: "Mconv1_stage3_L1"
+  top: "Mconv1_stage3_L1"
+}
+layer {
+  name: "Mconv1_stage3_L2"
+  type: "Convolution"
+  bottom: "concat_stage3"
+  top: "Mconv1_stage3_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage3_L2"
+  type: "ReLU"
+  bottom: "Mconv1_stage3_L2"
+  top: "Mconv1_stage3_L2"
+}
+layer {
+  name: "Mconv2_stage3_L1"
+  type: "Convolution"
+  bottom: "Mconv1_stage3_L1"
+  top: "Mconv2_stage3_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage3_L1"
+  type: "ReLU"
+  bottom: "Mconv2_stage3_L1"
+  top: "Mconv2_stage3_L1"
+}
+layer {
+  name: "Mconv2_stage3_L2"
+  type: "Convolution"
+  bottom: "Mconv1_stage3_L2"
+  top: "Mconv2_stage3_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage3_L2"
+  type: "ReLU"
+  bottom: "Mconv2_stage3_L2"
+  top: "Mconv2_stage3_L2"
+}
+layer {
+  name: "Mconv3_stage3_L1"
+  type: "Convolution"
+  bottom: "Mconv2_stage3_L1"
+  top: "Mconv3_stage3_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage3_L1"
+  type: "ReLU"
+  bottom: "Mconv3_stage3_L1"
+  top: "Mconv3_stage3_L1"
+}
+layer {
+  name: "Mconv3_stage3_L2"
+  type: "Convolution"
+  bottom: "Mconv2_stage3_L2"
+  top: "Mconv3_stage3_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage3_L2"
+  type: "ReLU"
+  bottom: "Mconv3_stage3_L2"
+  top: "Mconv3_stage3_L2"
+}
+layer {
+  name: "Mconv4_stage3_L1"
+  type: "Convolution"
+  bottom: "Mconv3_stage3_L1"
+  top: "Mconv4_stage3_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage3_L1"
+  type: "ReLU"
+  bottom: "Mconv4_stage3_L1"
+  top: "Mconv4_stage3_L1"
+}
+layer {
+  name: "Mconv4_stage3_L2"
+  type: "Convolution"
+  bottom: "Mconv3_stage3_L2"
+  top: "Mconv4_stage3_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage3_L2"
+  type: "ReLU"
+  bottom: "Mconv4_stage3_L2"
+  top: "Mconv4_stage3_L2"
+}
+layer {
+  name: "Mconv5_stage3_L1"
+  type: "Convolution"
+  bottom: "Mconv4_stage3_L1"
+  top: "Mconv5_stage3_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage3_L1"
+  type: "ReLU"
+  bottom: "Mconv5_stage3_L1"
+  top: "Mconv5_stage3_L1"
+}
+layer {
+  name: "Mconv5_stage3_L2"
+  type: "Convolution"
+  bottom: "Mconv4_stage3_L2"
+  top: "Mconv5_stage3_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage3_L2"
+  type: "ReLU"
+  bottom: "Mconv5_stage3_L2"
+  top: "Mconv5_stage3_L2"
+}
+layer {
+  name: "Mconv6_stage3_L1"
+  type: "Convolution"
+  bottom: "Mconv5_stage3_L1"
+  top: "Mconv6_stage3_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage3_L1"
+  type: "ReLU"
+  bottom: "Mconv6_stage3_L1"
+  top: "Mconv6_stage3_L1"
+}
+layer {
+  name: "Mconv6_stage3_L2"
+  type: "Convolution"
+  bottom: "Mconv5_stage3_L2"
+  top: "Mconv6_stage3_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage3_L2"
+  type: "ReLU"
+  bottom: "Mconv6_stage3_L2"
+  top: "Mconv6_stage3_L2"
+}
+layer {
+  name: "Mconv7_stage3_L1"
+  type: "Convolution"
+  bottom: "Mconv6_stage3_L1"
+  top: "Mconv7_stage3_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 38
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mconv7_stage3_L2"
+  type: "Convolution"
+  bottom: "Mconv6_stage3_L2"
+  top: "Mconv7_stage3_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 19
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "concat_stage4"
+  type: "Concat"
+  bottom: "Mconv7_stage3_L1"
+  bottom: "Mconv7_stage3_L2"
+  bottom: "conv4_4_CPM"
+  top: "concat_stage4"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage4_L1"
+  type: "Convolution"
+  bottom: "concat_stage4"
+  top: "Mconv1_stage4_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage4_L1"
+  type: "ReLU"
+  bottom: "Mconv1_stage4_L1"
+  top: "Mconv1_stage4_L1"
+}
+layer {
+  name: "Mconv1_stage4_L2"
+  type: "Convolution"
+  bottom: "concat_stage4"
+  top: "Mconv1_stage4_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage4_L2"
+  type: "ReLU"
+  bottom: "Mconv1_stage4_L2"
+  top: "Mconv1_stage4_L2"
+}
+layer {
+  name: "Mconv2_stage4_L1"
+  type: "Convolution"
+  bottom: "Mconv1_stage4_L1"
+  top: "Mconv2_stage4_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage4_L1"
+  type: "ReLU"
+  bottom: "Mconv2_stage4_L1"
+  top: "Mconv2_stage4_L1"
+}
+layer {
+  name: "Mconv2_stage4_L2"
+  type: "Convolution"
+  bottom: "Mconv1_stage4_L2"
+  top: "Mconv2_stage4_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage4_L2"
+  type: "ReLU"
+  bottom: "Mconv2_stage4_L2"
+  top: "Mconv2_stage4_L2"
+}
+layer {
+  name: "Mconv3_stage4_L1"
+  type: "Convolution"
+  bottom: "Mconv2_stage4_L1"
+  top: "Mconv3_stage4_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage4_L1"
+  type: "ReLU"
+  bottom: "Mconv3_stage4_L1"
+  top: "Mconv3_stage4_L1"
+}
+layer {
+  name: "Mconv3_stage4_L2"
+  type: "Convolution"
+  bottom: "Mconv2_stage4_L2"
+  top: "Mconv3_stage4_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage4_L2"
+  type: "ReLU"
+  bottom: "Mconv3_stage4_L2"
+  top: "Mconv3_stage4_L2"
+}
+layer {
+  name: "Mconv4_stage4_L1"
+  type: "Convolution"
+  bottom: "Mconv3_stage4_L1"
+  top: "Mconv4_stage4_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage4_L1"
+  type: "ReLU"
+  bottom: "Mconv4_stage4_L1"
+  top: "Mconv4_stage4_L1"
+}
+layer {
+  name: "Mconv4_stage4_L2"
+  type: "Convolution"
+  bottom: "Mconv3_stage4_L2"
+  top: "Mconv4_stage4_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage4_L2"
+  type: "ReLU"
+  bottom: "Mconv4_stage4_L2"
+  top: "Mconv4_stage4_L2"
+}
+layer {
+  name: "Mconv5_stage4_L1"
+  type: "Convolution"
+  bottom: "Mconv4_stage4_L1"
+  top: "Mconv5_stage4_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage4_L1"
+  type: "ReLU"
+  bottom: "Mconv5_stage4_L1"
+  top: "Mconv5_stage4_L1"
+}
+layer {
+  name: "Mconv5_stage4_L2"
+  type: "Convolution"
+  bottom: "Mconv4_stage4_L2"
+  top: "Mconv5_stage4_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage4_L2"
+  type: "ReLU"
+  bottom: "Mconv5_stage4_L2"
+  top: "Mconv5_stage4_L2"
+}
+layer {
+  name: "Mconv6_stage4_L1"
+  type: "Convolution"
+  bottom: "Mconv5_stage4_L1"
+  top: "Mconv6_stage4_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage4_L1"
+  type: "ReLU"
+  bottom: "Mconv6_stage4_L1"
+  top: "Mconv6_stage4_L1"
+}
+layer {
+  name: "Mconv6_stage4_L2"
+  type: "Convolution"
+  bottom: "Mconv5_stage4_L2"
+  top: "Mconv6_stage4_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage4_L2"
+  type: "ReLU"
+  bottom: "Mconv6_stage4_L2"
+  top: "Mconv6_stage4_L2"
+}
+layer {
+  name: "Mconv7_stage4_L1"
+  type: "Convolution"
+  bottom: "Mconv6_stage4_L1"
+  top: "Mconv7_stage4_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 38
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mconv7_stage4_L2"
+  type: "Convolution"
+  bottom: "Mconv6_stage4_L2"
+  top: "Mconv7_stage4_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 19
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "concat_stage5"
+  type: "Concat"
+  bottom: "Mconv7_stage4_L1"
+  bottom: "Mconv7_stage4_L2"
+  bottom: "conv4_4_CPM"
+  top: "concat_stage5"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage5_L1"
+  type: "Convolution"
+  bottom: "concat_stage5"
+  top: "Mconv1_stage5_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage5_L1"
+  type: "ReLU"
+  bottom: "Mconv1_stage5_L1"
+  top: "Mconv1_stage5_L1"
+}
+layer {
+  name: "Mconv1_stage5_L2"
+  type: "Convolution"
+  bottom: "concat_stage5"
+  top: "Mconv1_stage5_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage5_L2"
+  type: "ReLU"
+  bottom: "Mconv1_stage5_L2"
+  top: "Mconv1_stage5_L2"
+}
+layer {
+  name: "Mconv2_stage5_L1"
+  type: "Convolution"
+  bottom: "Mconv1_stage5_L1"
+  top: "Mconv2_stage5_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage5_L1"
+  type: "ReLU"
+  bottom: "Mconv2_stage5_L1"
+  top: "Mconv2_stage5_L1"
+}
+layer {
+  name: "Mconv2_stage5_L2"
+  type: "Convolution"
+  bottom: "Mconv1_stage5_L2"
+  top: "Mconv2_stage5_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage5_L2"
+  type: "ReLU"
+  bottom: "Mconv2_stage5_L2"
+  top: "Mconv2_stage5_L2"
+}
+layer {
+  name: "Mconv3_stage5_L1"
+  type: "Convolution"
+  bottom: "Mconv2_stage5_L1"
+  top: "Mconv3_stage5_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage5_L1"
+  type: "ReLU"
+  bottom: "Mconv3_stage5_L1"
+  top: "Mconv3_stage5_L1"
+}
+layer {
+  name: "Mconv3_stage5_L2"
+  type: "Convolution"
+  bottom: "Mconv2_stage5_L2"
+  top: "Mconv3_stage5_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage5_L2"
+  type: "ReLU"
+  bottom: "Mconv3_stage5_L2"
+  top: "Mconv3_stage5_L2"
+}
+layer {
+  name: "Mconv4_stage5_L1"
+  type: "Convolution"
+  bottom: "Mconv3_stage5_L1"
+  top: "Mconv4_stage5_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage5_L1"
+  type: "ReLU"
+  bottom: "Mconv4_stage5_L1"
+  top: "Mconv4_stage5_L1"
+}
+layer {
+  name: "Mconv4_stage5_L2"
+  type: "Convolution"
+  bottom: "Mconv3_stage5_L2"
+  top: "Mconv4_stage5_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage5_L2"
+  type: "ReLU"
+  bottom: "Mconv4_stage5_L2"
+  top: "Mconv4_stage5_L2"
+}
+layer {
+  name: "Mconv5_stage5_L1"
+  type: "Convolution"
+  bottom: "Mconv4_stage5_L1"
+  top: "Mconv5_stage5_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage5_L1"
+  type: "ReLU"
+  bottom: "Mconv5_stage5_L1"
+  top: "Mconv5_stage5_L1"
+}
+layer {
+  name: "Mconv5_stage5_L2"
+  type: "Convolution"
+  bottom: "Mconv4_stage5_L2"
+  top: "Mconv5_stage5_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage5_L2"
+  type: "ReLU"
+  bottom: "Mconv5_stage5_L2"
+  top: "Mconv5_stage5_L2"
+}
+layer {
+  name: "Mconv6_stage5_L1"
+  type: "Convolution"
+  bottom: "Mconv5_stage5_L1"
+  top: "Mconv6_stage5_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage5_L1"
+  type: "ReLU"
+  bottom: "Mconv6_stage5_L1"
+  top: "Mconv6_stage5_L1"
+}
+layer {
+  name: "Mconv6_stage5_L2"
+  type: "Convolution"
+  bottom: "Mconv5_stage5_L2"
+  top: "Mconv6_stage5_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage5_L2"
+  type: "ReLU"
+  bottom: "Mconv6_stage5_L2"
+  top: "Mconv6_stage5_L2"
+}
+layer {
+  name: "Mconv7_stage5_L1"
+  type: "Convolution"
+  bottom: "Mconv6_stage5_L1"
+  top: "Mconv7_stage5_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 38
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mconv7_stage5_L2"
+  type: "Convolution"
+  bottom: "Mconv6_stage5_L2"
+  top: "Mconv7_stage5_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 19
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "concat_stage6"
+  type: "Concat"
+  bottom: "Mconv7_stage5_L1"
+  bottom: "Mconv7_stage5_L2"
+  bottom: "conv4_4_CPM"
+  top: "concat_stage6"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage6_L1"
+  type: "Convolution"
+  bottom: "concat_stage6"
+  top: "Mconv1_stage6_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage6_L1"
+  type: "ReLU"
+  bottom: "Mconv1_stage6_L1"
+  top: "Mconv1_stage6_L1"
+}
+layer {
+  name: "Mconv1_stage6_L2"
+  type: "Convolution"
+  bottom: "concat_stage6"
+  top: "Mconv1_stage6_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu1_stage6_L2"
+  type: "ReLU"
+  bottom: "Mconv1_stage6_L2"
+  top: "Mconv1_stage6_L2"
+}
+layer {
+  name: "Mconv2_stage6_L1"
+  type: "Convolution"
+  bottom: "Mconv1_stage6_L1"
+  top: "Mconv2_stage6_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage6_L1"
+  type: "ReLU"
+  bottom: "Mconv2_stage6_L1"
+  top: "Mconv2_stage6_L1"
+}
+layer {
+  name: "Mconv2_stage6_L2"
+  type: "Convolution"
+  bottom: "Mconv1_stage6_L2"
+  top: "Mconv2_stage6_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu2_stage6_L2"
+  type: "ReLU"
+  bottom: "Mconv2_stage6_L2"
+  top: "Mconv2_stage6_L2"
+}
+layer {
+  name: "Mconv3_stage6_L1"
+  type: "Convolution"
+  bottom: "Mconv2_stage6_L1"
+  top: "Mconv3_stage6_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage6_L1"
+  type: "ReLU"
+  bottom: "Mconv3_stage6_L1"
+  top: "Mconv3_stage6_L1"
+}
+layer {
+  name: "Mconv3_stage6_L2"
+  type: "Convolution"
+  bottom: "Mconv2_stage6_L2"
+  top: "Mconv3_stage6_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu3_stage6_L2"
+  type: "ReLU"
+  bottom: "Mconv3_stage6_L2"
+  top: "Mconv3_stage6_L2"
+}
+layer {
+  name: "Mconv4_stage6_L1"
+  type: "Convolution"
+  bottom: "Mconv3_stage6_L1"
+  top: "Mconv4_stage6_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage6_L1"
+  type: "ReLU"
+  bottom: "Mconv4_stage6_L1"
+  top: "Mconv4_stage6_L1"
+}
+layer {
+  name: "Mconv4_stage6_L2"
+  type: "Convolution"
+  bottom: "Mconv3_stage6_L2"
+  top: "Mconv4_stage6_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu4_stage6_L2"
+  type: "ReLU"
+  bottom: "Mconv4_stage6_L2"
+  top: "Mconv4_stage6_L2"
+}
+layer {
+  name: "Mconv5_stage6_L1"
+  type: "Convolution"
+  bottom: "Mconv4_stage6_L1"
+  top: "Mconv5_stage6_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage6_L1"
+  type: "ReLU"
+  bottom: "Mconv5_stage6_L1"
+  top: "Mconv5_stage6_L1"
+}
+layer {
+  name: "Mconv5_stage6_L2"
+  type: "Convolution"
+  bottom: "Mconv4_stage6_L2"
+  top: "Mconv5_stage6_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu5_stage6_L2"
+  type: "ReLU"
+  bottom: "Mconv5_stage6_L2"
+  top: "Mconv5_stage6_L2"
+}
+layer {
+  name: "Mconv6_stage6_L1"
+  type: "Convolution"
+  bottom: "Mconv5_stage6_L1"
+  top: "Mconv6_stage6_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage6_L1"
+  type: "ReLU"
+  bottom: "Mconv6_stage6_L1"
+  top: "Mconv6_stage6_L1"
+}
+layer {
+  name: "Mconv6_stage6_L2"
+  type: "Convolution"
+  bottom: "Mconv5_stage6_L2"
+  top: "Mconv6_stage6_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mrelu6_stage6_L2"
+  type: "ReLU"
+  bottom: "Mconv6_stage6_L2"
+  top: "Mconv6_stage6_L2"
+}
+layer {
+  name: "Mconv7_stage6_L1"
+  type: "Convolution"
+  bottom: "Mconv6_stage6_L1"
+  top: "Mconv7_stage6_L1"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 38
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "Mconv7_stage6_L2"
+  type: "Convolution"
+  bottom: "Mconv6_stage6_L2"
+  top: "Mconv7_stage6_L2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 19
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+  }
+}
+layer {
+  name: "concat_stage7"
+  type: "Concat"
+  bottom: "Mconv7_stage6_L2"
+  bottom: "Mconv7_stage6_L1"
+  # top: "concat_stage7"
+  top: "net_output"
+  concat_param {
+    axis: 1
+  }
+}

model/hand_pose_deploy.prototxt

@@ -0,0 +1,1756 @@
+input: "image"
+input_dim: 1 # Original: 2
+input_dim: 3 # It crashes if not left to 3
+input_dim: 1 # Original: 368
+input_dim: 1 # Original: 368
+layer {
+  name: "conv1_1"
+  type: "Convolution"
+  bottom: "image"
+  top: "conv1_1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 64
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu1_1"
+  type: "ReLU"
+  bottom: "conv1_1"
+  top: "conv1_1"
+}
+layer {
+  name: "conv1_2"
+  type: "Convolution"
+  bottom: "conv1_1"
+  top: "conv1_2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 64
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu1_2"
+  type: "ReLU"
+  bottom: "conv1_2"
+  top: "conv1_2"
+}
+layer {
+  name: "pool1_stage1"
+  type: "Pooling"
+  bottom: "conv1_2"
+  top: "pool1_stage1"
+  pooling_param {
+    pool: MAX
+    kernel_size: 2
+    stride: 2
+  }
+}
+layer {
+  name: "conv2_1"
+  type: "Convolution"
+  bottom: "pool1_stage1"
+  top: "conv2_1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu2_1"
+  type: "ReLU"
+  bottom: "conv2_1"
+  top: "conv2_1"
+}
+layer {
+  name: "conv2_2"
+  type: "Convolution"
+  bottom: "conv2_1"
+  top: "conv2_2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu2_2"
+  type: "ReLU"
+  bottom: "conv2_2"
+  top: "conv2_2"
+}
+layer {
+  name: "pool2_stage1"
+  type: "Pooling"
+  bottom: "conv2_2"
+  top: "pool2_stage1"
+  pooling_param {
+    pool: MAX
+    kernel_size: 2
+    stride: 2
+  }
+}
+layer {
+  name: "conv3_1"
+  type: "Convolution"
+  bottom: "pool2_stage1"
+  top: "conv3_1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu3_1"
+  type: "ReLU"
+  bottom: "conv3_1"
+  top: "conv3_1"
+}
+layer {
+  name: "conv3_2"
+  type: "Convolution"
+  bottom: "conv3_1"
+  top: "conv3_2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu3_2"
+  type: "ReLU"
+  bottom: "conv3_2"
+  top: "conv3_2"
+}
+layer {
+  name: "conv3_3"
+  type: "Convolution"
+  bottom: "conv3_2"
+  top: "conv3_3"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu3_3"
+  type: "ReLU"
+  bottom: "conv3_3"
+  top: "conv3_3"
+}
+layer {
+  name: "conv3_4"
+  type: "Convolution"
+  bottom: "conv3_3"
+  top: "conv3_4"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 256
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu3_4"
+  type: "ReLU"
+  bottom: "conv3_4"
+  top: "conv3_4"
+}
+layer {
+  name: "pool3_stage1"
+  type: "Pooling"
+  bottom: "conv3_4"
+  top: "pool3_stage1"
+  pooling_param {
+    pool: MAX
+    kernel_size: 2
+    stride: 2
+  }
+}
+layer {
+  name: "conv4_1"
+  type: "Convolution"
+  bottom: "pool3_stage1"
+  top: "conv4_1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu4_1"
+  type: "ReLU"
+  bottom: "conv4_1"
+  top: "conv4_1"
+}
+layer {
+  name: "conv4_2"
+  type: "Convolution"
+  bottom: "conv4_1"
+  top: "conv4_2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu4_2"
+  type: "ReLU"
+  bottom: "conv4_2"
+  top: "conv4_2"
+}
+layer {
+  name: "conv4_3"
+  type: "Convolution"
+  bottom: "conv4_2"
+  top: "conv4_3"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu4_3"
+  type: "ReLU"
+  bottom: "conv4_3"
+  top: "conv4_3"
+}
+layer {
+  name: "conv4_4"
+  type: "Convolution"
+  bottom: "conv4_3"
+  top: "conv4_4"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu4_4"
+  type: "ReLU"
+  bottom: "conv4_4"
+  top: "conv4_4"
+}
+layer {
+  name: "conv5_1"
+  type: "Convolution"
+  bottom: "conv4_4"
+  top: "conv5_1"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu5_1"
+  type: "ReLU"
+  bottom: "conv5_1"
+  top: "conv5_1"
+}
+layer {
+  name: "conv5_2"
+  type: "Convolution"
+  bottom: "conv5_1"
+  top: "conv5_2"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "xavier"
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu5_2"
+  type: "ReLU"
+  bottom: "conv5_2"
+  top: "conv5_2"
+}
+layer {
+  name: "conv5_3_CPM"
+  type: "Convolution"
+  bottom: "conv5_2"
+  top: "conv5_3_CPM"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 1
+    kernel_size: 3
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu5_4_stage1_3"
+  type: "ReLU"
+  bottom: "conv5_3_CPM"
+  top: "conv5_3_CPM"
+}
+layer {
+  name: "conv6_1_CPM"
+  type: "Convolution"
+  bottom: "conv5_3_CPM"
+  top: "conv6_1_CPM"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 512
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "relu6_4_stage1_1"
+  type: "ReLU"
+  bottom: "conv6_1_CPM"
+  top: "conv6_1_CPM"
+}
+layer {
+  name: "conv6_2_CPM"
+  type: "Convolution"
+  bottom: "conv6_1_CPM"
+  top: "conv6_2_CPM"
+  param {
+    lr_mult: 1.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 2.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 22
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "concat_stage2"
+  type: "Concat"
+  bottom: "conv6_2_CPM"
+  bottom: "conv5_3_CPM"
+  top: "concat_stage2"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage2"
+  type: "Convolution"
+  bottom: "concat_stage2"
+  top: "Mconv1_stage2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_2_stage2_1"
+  type: "ReLU"
+  bottom: "Mconv1_stage2"
+  top: "Mconv1_stage2"
+}
+layer {
+  name: "Mconv2_stage2"
+  type: "Convolution"
+  bottom: "Mconv1_stage2"
+  top: "Mconv2_stage2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_3_stage2_2"
+  type: "ReLU"
+  bottom: "Mconv2_stage2"
+  top: "Mconv2_stage2"
+}
+layer {
+  name: "Mconv3_stage2"
+  type: "Convolution"
+  bottom: "Mconv2_stage2"
+  top: "Mconv3_stage2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_4_stage2_3"
+  type: "ReLU"
+  bottom: "Mconv3_stage2"
+  top: "Mconv3_stage2"
+}
+layer {
+  name: "Mconv4_stage2"
+  type: "Convolution"
+  bottom: "Mconv3_stage2"
+  top: "Mconv4_stage2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_5_stage2_4"
+  type: "ReLU"
+  bottom: "Mconv4_stage2"
+  top: "Mconv4_stage2"
+}
+layer {
+  name: "Mconv5_stage2"
+  type: "Convolution"
+  bottom: "Mconv4_stage2"
+  top: "Mconv5_stage2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_6_stage2_5"
+  type: "ReLU"
+  bottom: "Mconv5_stage2"
+  top: "Mconv5_stage2"
+}
+layer {
+  name: "Mconv6_stage2"
+  type: "Convolution"
+  bottom: "Mconv5_stage2"
+  top: "Mconv6_stage2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_7_stage2_6"
+  type: "ReLU"
+  bottom: "Mconv6_stage2"
+  top: "Mconv6_stage2"
+}
+layer {
+  name: "Mconv7_stage2"
+  type: "Convolution"
+  bottom: "Mconv6_stage2"
+  top: "Mconv7_stage2"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 22
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "concat_stage3"
+  type: "Concat"
+  bottom: "Mconv7_stage2"
+  bottom: "conv5_3_CPM"
+  top: "concat_stage3"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage3"
+  type: "Convolution"
+  bottom: "concat_stage3"
+  top: "Mconv1_stage3"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_2_stage3_1"
+  type: "ReLU"
+  bottom: "Mconv1_stage3"
+  top: "Mconv1_stage3"
+}
+layer {
+  name: "Mconv2_stage3"
+  type: "Convolution"
+  bottom: "Mconv1_stage3"
+  top: "Mconv2_stage3"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_3_stage3_2"
+  type: "ReLU"
+  bottom: "Mconv2_stage3"
+  top: "Mconv2_stage3"
+}
+layer {
+  name: "Mconv3_stage3"
+  type: "Convolution"
+  bottom: "Mconv2_stage3"
+  top: "Mconv3_stage3"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_4_stage3_3"
+  type: "ReLU"
+  bottom: "Mconv3_stage3"
+  top: "Mconv3_stage3"
+}
+layer {
+  name: "Mconv4_stage3"
+  type: "Convolution"
+  bottom: "Mconv3_stage3"
+  top: "Mconv4_stage3"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_5_stage3_4"
+  type: "ReLU"
+  bottom: "Mconv4_stage3"
+  top: "Mconv4_stage3"
+}
+layer {
+  name: "Mconv5_stage3"
+  type: "Convolution"
+  bottom: "Mconv4_stage3"
+  top: "Mconv5_stage3"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_6_stage3_5"
+  type: "ReLU"
+  bottom: "Mconv5_stage3"
+  top: "Mconv5_stage3"
+}
+layer {
+  name: "Mconv6_stage3"
+  type: "Convolution"
+  bottom: "Mconv5_stage3"
+  top: "Mconv6_stage3"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_7_stage3_6"
+  type: "ReLU"
+  bottom: "Mconv6_stage3"
+  top: "Mconv6_stage3"
+}
+layer {
+  name: "Mconv7_stage3"
+  type: "Convolution"
+  bottom: "Mconv6_stage3"
+  top: "Mconv7_stage3"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 22
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "concat_stage4"
+  type: "Concat"
+  bottom: "Mconv7_stage3"
+  bottom: "conv5_3_CPM"
+  top: "concat_stage4"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage4"
+  type: "Convolution"
+  bottom: "concat_stage4"
+  top: "Mconv1_stage4"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_2_stage4_1"
+  type: "ReLU"
+  bottom: "Mconv1_stage4"
+  top: "Mconv1_stage4"
+}
+layer {
+  name: "Mconv2_stage4"
+  type: "Convolution"
+  bottom: "Mconv1_stage4"
+  top: "Mconv2_stage4"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_3_stage4_2"
+  type: "ReLU"
+  bottom: "Mconv2_stage4"
+  top: "Mconv2_stage4"
+}
+layer {
+  name: "Mconv3_stage4"
+  type: "Convolution"
+  bottom: "Mconv2_stage4"
+  top: "Mconv3_stage4"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_4_stage4_3"
+  type: "ReLU"
+  bottom: "Mconv3_stage4"
+  top: "Mconv3_stage4"
+}
+layer {
+  name: "Mconv4_stage4"
+  type: "Convolution"
+  bottom: "Mconv3_stage4"
+  top: "Mconv4_stage4"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_5_stage4_4"
+  type: "ReLU"
+  bottom: "Mconv4_stage4"
+  top: "Mconv4_stage4"
+}
+layer {
+  name: "Mconv5_stage4"
+  type: "Convolution"
+  bottom: "Mconv4_stage4"
+  top: "Mconv5_stage4"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_6_stage4_5"
+  type: "ReLU"
+  bottom: "Mconv5_stage4"
+  top: "Mconv5_stage4"
+}
+layer {
+  name: "Mconv6_stage4"
+  type: "Convolution"
+  bottom: "Mconv5_stage4"
+  top: "Mconv6_stage4"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_7_stage4_6"
+  type: "ReLU"
+  bottom: "Mconv6_stage4"
+  top: "Mconv6_stage4"
+}
+layer {
+  name: "Mconv7_stage4"
+  type: "Convolution"
+  bottom: "Mconv6_stage4"
+  top: "Mconv7_stage4"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 22
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "concat_stage5"
+  type: "Concat"
+  bottom: "Mconv7_stage4"
+  bottom: "conv5_3_CPM"
+  top: "concat_stage5"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage5"
+  type: "Convolution"
+  bottom: "concat_stage5"
+  top: "Mconv1_stage5"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_2_stage5_1"
+  type: "ReLU"
+  bottom: "Mconv1_stage5"
+  top: "Mconv1_stage5"
+}
+layer {
+  name: "Mconv2_stage5"
+  type: "Convolution"
+  bottom: "Mconv1_stage5"
+  top: "Mconv2_stage5"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_3_stage5_2"
+  type: "ReLU"
+  bottom: "Mconv2_stage5"
+  top: "Mconv2_stage5"
+}
+layer {
+  name: "Mconv3_stage5"
+  type: "Convolution"
+  bottom: "Mconv2_stage5"
+  top: "Mconv3_stage5"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_4_stage5_3"
+  type: "ReLU"
+  bottom: "Mconv3_stage5"
+  top: "Mconv3_stage5"
+}
+layer {
+  name: "Mconv4_stage5"
+  type: "Convolution"
+  bottom: "Mconv3_stage5"
+  top: "Mconv4_stage5"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_5_stage5_4"
+  type: "ReLU"
+  bottom: "Mconv4_stage5"
+  top: "Mconv4_stage5"
+}
+layer {
+  name: "Mconv5_stage5"
+  type: "Convolution"
+  bottom: "Mconv4_stage5"
+  top: "Mconv5_stage5"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_6_stage5_5"
+  type: "ReLU"
+  bottom: "Mconv5_stage5"
+  top: "Mconv5_stage5"
+}
+layer {
+  name: "Mconv6_stage5"
+  type: "Convolution"
+  bottom: "Mconv5_stage5"
+  top: "Mconv6_stage5"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_7_stage5_6"
+  type: "ReLU"
+  bottom: "Mconv6_stage5"
+  top: "Mconv6_stage5"
+}
+layer {
+  name: "Mconv7_stage5"
+  type: "Convolution"
+  bottom: "Mconv6_stage5"
+  top: "Mconv7_stage5"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 22
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "concat_stage6"
+  type: "Concat"
+  bottom: "Mconv7_stage5"
+  bottom: "conv5_3_CPM"
+  top: "concat_stage6"
+  concat_param {
+    axis: 1
+  }
+}
+layer {
+  name: "Mconv1_stage6"
+  type: "Convolution"
+  bottom: "concat_stage6"
+  top: "Mconv1_stage6"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_2_stage6_1"
+  type: "ReLU"
+  bottom: "Mconv1_stage6"
+  top: "Mconv1_stage6"
+}
+layer {
+  name: "Mconv2_stage6"
+  type: "Convolution"
+  bottom: "Mconv1_stage6"
+  top: "Mconv2_stage6"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_3_stage6_2"
+  type: "ReLU"
+  bottom: "Mconv2_stage6"
+  top: "Mconv2_stage6"
+}
+layer {
+  name: "Mconv3_stage6"
+  type: "Convolution"
+  bottom: "Mconv2_stage6"
+  top: "Mconv3_stage6"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_4_stage6_3"
+  type: "ReLU"
+  bottom: "Mconv3_stage6"
+  top: "Mconv3_stage6"
+}
+layer {
+  name: "Mconv4_stage6"
+  type: "Convolution"
+  bottom: "Mconv3_stage6"
+  top: "Mconv4_stage6"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_5_stage6_4"
+  type: "ReLU"
+  bottom: "Mconv4_stage6"
+  top: "Mconv4_stage6"
+}
+layer {
+  name: "Mconv5_stage6"
+  type: "Convolution"
+  bottom: "Mconv4_stage6"
+  top: "Mconv5_stage6"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 3
+    kernel_size: 7
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_6_stage6_5"
+  type: "ReLU"
+  bottom: "Mconv5_stage6"
+  top: "Mconv5_stage6"
+}
+layer {
+  name: "Mconv6_stage6"
+  type: "Convolution"
+  bottom: "Mconv5_stage6"
+  top: "Mconv6_stage6"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 128
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+layer {
+  name: "Mrelu1_7_stage6_6"
+  type: "ReLU"
+  bottom: "Mconv6_stage6"
+  top: "Mconv6_stage6"
+}
+layer {
+  name: "Mconv7_stage6"
+  type: "Convolution"
+  bottom: "Mconv6_stage6"
+#   top: "Mconv7_stage6"
+  top: "net_output"
+  param {
+    lr_mult: 4.0
+    decay_mult: 1
+  }
+  param {
+    lr_mult: 8.0
+    decay_mult: 0
+  }
+  convolution_param {
+    num_output: 22
+    pad: 0
+    kernel_size: 1
+    weight_filler {
+      type: "gaussian"
+      std: 0.01
+    }
+    bias_filler {
+      type: "constant"
+    }
+    dilation: 1
+  }
+}
+

requirements.txt

@@ -0,0 +1,6 @@
+numpy
+matplotlib
+opencv-python
+scipy
+scikit-image
+tqdm

src/body.py

@@ -0,0 +1,218 @@
+import cv2
+import numpy as np
+import math
+import time
+from scipy.ndimage.filters import gaussian_filter
+import matplotlib.pyplot as plt
+import matplotlib
+import torch
+from torchvision import transforms
+
+from src import util
+from src.model import bodypose_model
+
+class Body(object):
+    def __init__(self, model_path):
+        self.model = bodypose_model()
+        if torch.cuda.is_available():
+            self.model = self.model.cuda()
+        model_dict = util.transfer(self.model, torch.load(model_path))
+        self.model.load_state_dict(model_dict)
+        self.model.eval()
+
+    def __call__(self, oriImg):
+        # scale_search = [0.5, 1.0, 1.5, 2.0]
+        scale_search = [0.5]
+        boxsize = 368
+        stride = 8
+        padValue = 128
+        thre1 = 0.1
+        thre2 = 0.05
+        multiplier = [x * boxsize / oriImg.shape[0] for x in scale_search]
+        heatmap_avg = np.zeros((oriImg.shape[0], oriImg.shape[1], 19))
+        paf_avg = np.zeros((oriImg.shape[0], oriImg.shape[1], 38))
+
+        for m in range(len(multiplier)):
+            scale = multiplier[m]
+            imageToTest = cv2.resize(oriImg, (0, 0), fx=scale, fy=scale, interpolation=cv2.INTER_CUBIC)
+            imageToTest_padded, pad = util.padRightDownCorner(imageToTest, stride, padValue)
+            im = np.transpose(np.float32(imageToTest_padded[:, :, :, np.newaxis]), (3, 2, 0, 1)) / 256 - 0.5
+            im = np.ascontiguousarray(im)
+
+            data = torch.from_numpy(im).float()
+            if torch.cuda.is_available():
+                data = data.cuda()
+            # data = data.permute([2, 0, 1]).unsqueeze(0).float()
+            with torch.no_grad():
+                Mconv7_stage6_L1, Mconv7_stage6_L2 = self.model(data)
+            Mconv7_stage6_L1 = Mconv7_stage6_L1.cpu().numpy()
+            Mconv7_stage6_L2 = Mconv7_stage6_L2.cpu().numpy()
+
+            # extract outputs, resize, and remove padding
+            # heatmap = np.transpose(np.squeeze(net.blobs[output_blobs.keys()[1]].data), (1, 2, 0))  # output 1 is heatmaps
+            heatmap = np.transpose(np.squeeze(Mconv7_stage6_L2), (1, 2, 0))  # output 1 is heatmaps
+            heatmap = cv2.resize(heatmap, (0, 0), fx=stride, fy=stride, interpolation=cv2.INTER_CUBIC)
+            heatmap = heatmap[:imageToTest_padded.shape[0] - pad[2], :imageToTest_padded.shape[1] - pad[3], :]
+            heatmap = cv2.resize(heatmap, (oriImg.shape[1], oriImg.shape[0]), interpolation=cv2.INTER_CUBIC)
+
+            # paf = np.transpose(np.squeeze(net.blobs[output_blobs.keys()[0]].data), (1, 2, 0))  # output 0 is PAFs
+            paf = np.transpose(np.squeeze(Mconv7_stage6_L1), (1, 2, 0))  # output 0 is PAFs
+            paf = cv2.resize(paf, (0, 0), fx=stride, fy=stride, interpolation=cv2.INTER_CUBIC)
+            paf = paf[:imageToTest_padded.shape[0] - pad[2], :imageToTest_padded.shape[1] - pad[3], :]
+            paf = cv2.resize(paf, (oriImg.shape[1], oriImg.shape[0]), interpolation=cv2.INTER_CUBIC)
+
+            heatmap_avg += heatmap_avg + heatmap / len(multiplier)
+            paf_avg += + paf / len(multiplier)
+
+        all_peaks = []
+        peak_counter = 0
+
+        for part in range(18):
+            map_ori = heatmap_avg[:, :, part]
+            one_heatmap = gaussian_filter(map_ori, sigma=3)
+
+            map_left = np.zeros(one_heatmap.shape)
+            map_left[1:, :] = one_heatmap[:-1, :]
+            map_right = np.zeros(one_heatmap.shape)
+            map_right[:-1, :] = one_heatmap[1:, :]
+            map_up = np.zeros(one_heatmap.shape)
+            map_up[:, 1:] = one_heatmap[:, :-1]
+            map_down = np.zeros(one_heatmap.shape)
+            map_down[:, :-1] = one_heatmap[:, 1:]
+
+            peaks_binary = np.logical_and.reduce(
+                (one_heatmap >= map_left, one_heatmap >= map_right, one_heatmap >= map_up, one_heatmap >= map_down, one_heatmap > thre1))
+            peaks = list(zip(np.nonzero(peaks_binary)[1], np.nonzero(peaks_binary)[0]))  # note reverse
+            peaks_with_score = [x + (map_ori[x[1], x[0]],) for x in peaks]
+            peak_id = range(peak_counter, peak_counter + len(peaks))
+            peaks_with_score_and_id = [peaks_with_score[i] + (peak_id[i],) for i in range(len(peak_id))]
+
+            all_peaks.append(peaks_with_score_and_id)
+            peak_counter += len(peaks)
+
+        # find connection in the specified sequence, center 29 is in the position 15
+        limbSeq = [[2, 3], [2, 6], [3, 4], [4, 5], [6, 7], [7, 8], [2, 9], [9, 10], \
+                   [10, 11], [2, 12], [12, 13], [13, 14], [2, 1], [1, 15], [15, 17], \
+                   [1, 16], [16, 18], [3, 17], [6, 18]]
+        # the middle joints heatmap correpondence
+        mapIdx = [[31, 32], [39, 40], [33, 34], [35, 36], [41, 42], [43, 44], [19, 20], [21, 22], \
+                  [23, 24], [25, 26], [27, 28], [29, 30], [47, 48], [49, 50], [53, 54], [51, 52], \
+                  [55, 56], [37, 38], [45, 46]]
+
+        connection_all = []
+        special_k = []
+        mid_num = 10
+
+        for k in range(len(mapIdx)):
+            score_mid = paf_avg[:, :, [x - 19 for x in mapIdx[k]]]
+            candA = all_peaks[limbSeq[k][0] - 1]
+            candB = all_peaks[limbSeq[k][1] - 1]
+            nA = len(candA)
+            nB = len(candB)
+            indexA, indexB = limbSeq[k]
+            if (nA != 0 and nB != 0):
+                connection_candidate = []
+                for i in range(nA):
+                    for j in range(nB):
+                        vec = np.subtract(candB[j][:2], candA[i][:2])
+                        norm = math.sqrt(vec[0] * vec[0] + vec[1] * vec[1])
+                        norm = max(0.001, norm)
+                        vec = np.divide(vec, norm)
+
+                        startend = list(zip(np.linspace(candA[i][0], candB[j][0], num=mid_num), \
+                                            np.linspace(candA[i][1], candB[j][1], num=mid_num)))
+
+                        vec_x = np.array([score_mid[int(round(startend[I][1])), int(round(startend[I][0])), 0] \
+                                          for I in range(len(startend))])
+                        vec_y = np.array([score_mid[int(round(startend[I][1])), int(round(startend[I][0])), 1] \
+                                          for I in range(len(startend))])
+
+                        score_midpts = np.multiply(vec_x, vec[0]) + np.multiply(vec_y, vec[1])
+                        score_with_dist_prior = sum(score_midpts) / len(score_midpts) + min(
+                            0.5 * oriImg.shape[0] / norm - 1, 0)
+                        criterion1 = len(np.nonzero(score_midpts > thre2)[0]) > 0.8 * len(score_midpts)
+                        criterion2 = score_with_dist_prior > 0
+                        if criterion1 and criterion2:
+                            connection_candidate.append(
+                                [i, j, score_with_dist_prior, score_with_dist_prior + candA[i][2] + candB[j][2]])
+
+                connection_candidate = sorted(connection_candidate, key=lambda x: x[2], reverse=True)
+                connection = np.zeros((0, 5))
+                for c in range(len(connection_candidate)):
+                    i, j, s = connection_candidate[c][0:3]
+                    if (i not in connection[:, 3] and j not in connection[:, 4]):
+                        connection = np.vstack([connection, [candA[i][3], candB[j][3], s, i, j]])
+                        if (len(connection) >= min(nA, nB)):
+                            break
+
+                connection_all.append(connection)
+            else:
+                special_k.append(k)
+                connection_all.append([])
+
+        # last number in each row is the total parts number of that person
+        # the second last number in each row is the score of the overall configuration
+        subset = -1 * np.ones((0, 20))
+        candidate = np.array([item for sublist in all_peaks for item in sublist])
+
+        for k in range(len(mapIdx)):
+            if k not in special_k:
+                partAs = connection_all[k][:, 0]
+                partBs = connection_all[k][:, 1]
+                indexA, indexB = np.array(limbSeq[k]) - 1
+
+                for i in range(len(connection_all[k])):  # = 1:size(temp,1)
+                    found = 0
+                    subset_idx = [-1, -1]
+                    for j in range(len(subset)):  # 1:size(subset,1):
+                        if subset[j][indexA] == partAs[i] or subset[j][indexB] == partBs[i]:
+                            subset_idx[found] = j
+                            found += 1
+
+                    if found == 1:
+                        j = subset_idx[0]
+                        if subset[j][indexB] != partBs[i]:
+                            subset[j][indexB] = partBs[i]
+                            subset[j][-1] += 1
+                            subset[j][-2] += candidate[partBs[i].astype(int), 2] + connection_all[k][i][2]
+                    elif found == 2:  # if found 2 and disjoint, merge them
+                        j1, j2 = subset_idx
+                        membership = ((subset[j1] >= 0).astype(int) + (subset[j2] >= 0).astype(int))[:-2]
+                        if len(np.nonzero(membership == 2)[0]) == 0:  # merge
+                            subset[j1][:-2] += (subset[j2][:-2] + 1)
+                            subset[j1][-2:] += subset[j2][-2:]
+                            subset[j1][-2] += connection_all[k][i][2]
+                            subset = np.delete(subset, j2, 0)
+                        else:  # as like found == 1
+                            subset[j1][indexB] = partBs[i]
+                            subset[j1][-1] += 1
+                            subset[j1][-2] += candidate[partBs[i].astype(int), 2] + connection_all[k][i][2]
+
+                    # if find no partA in the subset, create a new subset
+                    elif not found and k < 17:
+                        row = -1 * np.ones(20)
+                        row[indexA] = partAs[i]
+                        row[indexB] = partBs[i]
+                        row[-1] = 2
+                        row[-2] = sum(candidate[connection_all[k][i, :2].astype(int), 2]) + connection_all[k][i][2]
+                        subset = np.vstack([subset, row])
+        # delete some rows of subset which has few parts occur
+        deleteIdx = []
+        for i in range(len(subset)):
+            if subset[i][-1] < 4 or subset[i][-2] / subset[i][-1] < 0.4:
+                deleteIdx.append(i)
+        subset = np.delete(subset, deleteIdx, axis=0)
+
+        # subset: n*20 array, 0-17 is the index in candidate, 18 is the total score, 19 is the total parts
+        # candidate: x, y, score, id
+        return candidate, subset
+
+if __name__ == "__main__":
+    body_estimation = Body('../model/body_pose_model.pth')
+
+    test_image = '../images/ski.jpg'
+    oriImg = cv2.imread(test_image)  # B,G,R order
+    candidate, subset = body_estimation(oriImg)
+    canvas = util.draw_bodypose(oriImg, candidate, subset)
+    plt.imshow(canvas[:, :, [2, 1, 0]])
+    plt.show()

src/hand.py

@@ -0,0 +1,85 @@
+import cv2
+import json
+import numpy as np
+import math
+import time
+from scipy.ndimage.filters import gaussian_filter
+import matplotlib.pyplot as plt
+import matplotlib
+import torch
+from skimage.measure import label
+
+from src.model import handpose_model
+from src import util
+
+class Hand(object):
+    def __init__(self, model_path):
+        self.model = handpose_model()
+        if torch.cuda.is_available():
+            self.model = self.model.cuda()
+        model_dict = util.transfer(self.model, torch.load(model_path))
+        self.model.load_state_dict(model_dict)
+        self.model.eval()
+
+    def __call__(self, oriImg):
+        scale_search = [0.5, 1.0, 1.5, 2.0]
+        # scale_search = [0.5]
+        boxsize = 368
+        stride = 8
+        padValue = 128
+        thre = 0.05
+        multiplier = [x * boxsize / oriImg.shape[0] for x in scale_search]
+        heatmap_avg = np.zeros((oriImg.shape[0], oriImg.shape[1], 22))
+        # paf_avg = np.zeros((oriImg.shape[0], oriImg.shape[1], 38))
+
+        for m in range(len(multiplier)):
+            scale = multiplier[m]
+            imageToTest = cv2.resize(oriImg, (0, 0), fx=scale, fy=scale, interpolation=cv2.INTER_CUBIC)
+            imageToTest_padded, pad = util.padRightDownCorner(imageToTest, stride, padValue)
+            im = np.transpose(np.float32(imageToTest_padded[:, :, :, np.newaxis]), (3, 2, 0, 1)) / 256 - 0.5
+            im = np.ascontiguousarray(im)
+
+            data = torch.from_numpy(im).float()
+            if torch.cuda.is_available():
+                data = data.cuda()
+            # data = data.permute([2, 0, 1]).unsqueeze(0).float()
+            with torch.no_grad():
+                output = self.model(data).cpu().numpy()
+                # output = self.model(data).numpy()q
+
+            # extract outputs, resize, and remove padding
+            heatmap = np.transpose(np.squeeze(output), (1, 2, 0))  # output 1 is heatmaps
+            heatmap = cv2.resize(heatmap, (0, 0), fx=stride, fy=stride, interpolation=cv2.INTER_CUBIC)
+            heatmap = heatmap[:imageToTest_padded.shape[0] - pad[2], :imageToTest_padded.shape[1] - pad[3], :]
+            heatmap = cv2.resize(heatmap, (oriImg.shape[1], oriImg.shape[0]), interpolation=cv2.INTER_CUBIC)
+
+            heatmap_avg += heatmap / len(multiplier)
+
+        all_peaks = []
+        for part in range(21):
+            map_ori = heatmap_avg[:, :, part]
+            one_heatmap = gaussian_filter(map_ori, sigma=3)
+            binary = np.ascontiguousarray(one_heatmap > thre, dtype=np.uint8)
+            # 全部小于阈值
+            if np.sum(binary) == 0:
+                all_peaks.append([0, 0])
+                continue
+            label_img, label_numbers = label(binary, return_num=True, connectivity=binary.ndim)
+            max_index = np.argmax([np.sum(map_ori[label_img == i]) for i in range(1, label_numbers + 1)]) + 1
+            label_img[label_img != max_index] = 0
+            map_ori[label_img == 0] = 0
+
+            y, x = util.npmax(map_ori)
+            all_peaks.append([x, y])
+        return np.array(all_peaks)
+
+if __name__ == "__main__":
+    hand_estimation = Hand('../model/hand_pose_model.pth')
+
+    # test_image = '../images/hand.jpg'
+    test_image = '../images/hand.jpg'
+    oriImg = cv2.imread(test_image)  # B,G,R order
+    peaks = hand_estimation(oriImg)
+    canvas = util.draw_handpose(oriImg, peaks, True)
+    cv2.imshow('', canvas)
+    cv2.waitKey(0)

src/hand_model_output_size.json

@@ -0,0 +1,992 @@
+{
+    "10":1,
+    "11":1,
+    "12":1,
+    "13":1,
+    "14":1,
+    "15":1,
+    "16":2,
+    "17":2,
+    "18":2,
+    "19":2,
+    "20":2,
+    "21":2,
+    "22":2,
+    "23":2,
+    "24":3,
+    "25":3,
+    "26":3,
+    "27":3,
+    "28":3,
+    "29":3,
+    "30":3,
+    "31":3,
+    "32":4,
+    "33":4,
+    "34":4,
+    "35":4,
+    "36":4,
+    "37":4,
+    "38":4,
+    "39":4,
+    "40":5,
+    "41":5,
+    "42":5,
+    "43":5,
+    "44":5,
+    "45":5,
+    "46":5,
+    "47":5,
+    "48":6,
+    "49":6,
+    "50":6,
+    "51":6,
+    "52":6,
+    "53":6,
+    "54":6,
+    "55":6,
+    "56":7,
+    "57":7,
+    "58":7,
+    "59":7,
+    "60":7,
+    "61":7,
+    "62":7,
+    "63":7,
+    "64":8,
+    "65":8,
+    "66":8,
+    "67":8,
+    "68":8,
+    "69":8,
+    "70":8,
+    "71":8,
+    "72":9,
+    "73":9,
+    "74":9,
+    "75":9,
+    "76":9,
+    "77":9,
+    "78":9,
+    "79":9,
+    "80":10,
+    "81":10,
+    "82":10,
+    "83":10,
+    "84":10,
+    "85":10,
+    "86":10,
+    "87":10,
+    "88":11,
+    "89":11,
+    "90":11,
+    "91":11,
+    "92":11,
+    "93":11,
+    "94":11,
+    "95":11,
+    "96":12,
+    "97":12,
+    "98":12,
+    "99":12,
+    "100":12,
+    "101":12,
+    "102":12,
+    "103":12,
+    "104":13,
+    "105":13,
+    "106":13,
+    "107":13,
+    "108":13,
+    "109":13,
+    "110":13,
+    "111":13,
+    "112":14,
+    "113":14,
+    "114":14,
+    "115":14,
+    "116":14,
+    "117":14,
+    "118":14,
+    "119":14,
+    "120":15,
+    "121":15,
+    "122":15,
+    "123":15,
+    "124":15,
+    "125":15,
+    "126":15,
+    "127":15,
+    "128":16,
+    "129":16,
+    "130":16,
+    "131":16,
+    "132":16,
+    "133":16,
+    "134":16,
+    "135":16,
+    "136":17,
+    "137":17,
+    "138":17,
+    "139":17,
+    "140":17,
+    "141":17,
+    "142":17,
+    "143":17,
+    "144":18,
+    "145":18,
+    "146":18,
+    "147":18,
+    "148":18,
+    "149":18,
+    "150":18,
+    "151":18,
+    "152":19,
+    "153":19,
+    "154":19,
+    "155":19,
+    "156":19,
+    "157":19,
+    "158":19,
+    "159":19,
+    "160":20,
+    "161":20,
+    "162":20,
+    "163":20,
+    "164":20,
+    "165":20,
+    "166":20,
+    "167":20,
+    "168":21,
+    "169":21,
+    "170":21,
+    "171":21,
+    "172":21,
+    "173":21,
+    "174":21,
+    "175":21,
+    "176":22,
+    "177":22,
+    "178":22,
+    "179":22,
+    "180":22,
+    "181":22,
+    "182":22,
+    "183":22,
+    "184":23,
+    "185":23,
+    "186":23,
+    "187":23,
+    "188":23,
+    "189":23,
+    "190":23,
+    "191":23,
+    "192":24,
+    "193":24,
+    "194":24,
+    "195":24,
+    "196":24,
+    "197":24,
+    "198":24,
+    "199":24,
+    "200":25,
+    "201":25,
+    "202":25,
+    "203":25,
+    "204":25,
+    "205":25,
+    "206":25,
+    "207":25,
+    "208":26,
+    "209":26,
+    "210":26,
+    "211":26,
+    "212":26,
+    "213":26,
+    "214":26,
+    "215":26,
+    "216":27,
+    "217":27,
+    "218":27,
+    "219":27,
+    "220":27,
+    "221":27,
+    "222":27,
+    "223":27,
+    "224":28,
+    "225":28,
+    "226":28,
+    "227":28,
+    "228":28,
+    "229":28,
+    "230":28,
+    "231":28,
+    "232":29,
+    "233":29,
+    "234":29,
+    "235":29,
+    "236":29,
+    "237":29,
+    "238":29,
+    "239":29,
+    "240":30,
+    "241":30,
+    "242":30,
+    "243":30,
+    "244":30,
+    "245":30,
+    "246":30,
+    "247":30,
+    "248":31,
+    "249":31,
+    "250":31,
+    "251":31,
+    "252":31,
+    "253":31,
+    "254":31,
+    "255":31,
+    "256":32,
+    "257":32,
+    "258":32,
+    "259":32,
+    "260":32,
+    "261":32,
+    "262":32,
+    "263":32,
+    "264":33,
+    "265":33,
+    "266":33,
+    "267":33,
+    "268":33,
+    "269":33,
+    "270":33,
+    "271":33,
+    "272":34,
+    "273":34,
+    "274":34,
+    "275":34,
+    "276":34,
+    "277":34,
+    "278":34,
+    "279":34,
+    "280":35,
+    "281":35,
+    "282":35,
+    "283":35,
+    "284":35,
+    "285":35,
+    "286":35,
+    "287":35,
+    "288":36,
+    "289":36,
+    "290":36,
+    "291":36,
+    "292":36,
+    "293":36,
+    "294":36,
+    "295":36,
+    "296":37,
+    "297":37,
+    "298":37,
+    "299":37,
+    "300":37,
+    "301":37,
+    "302":37,
+    "303":37,
+    "304":38,
+    "305":38,
+    "306":38,
+    "307":38,
+    "308":38,
+    "309":38,
+    "310":38,
+    "311":38,
+    "312":39,
+    "313":39,
+    "314":39,
+    "315":39,
+    "316":39,
+    "317":39,
+    "318":39,
+    "319":39,
+    "320":40,
+    "321":40,
+    "322":40,
+    "323":40,
+    "324":40,
+    "325":40,
+    "326":40,
+    "327":40,
+    "328":41,
+    "329":41,
+    "330":41,
+    "331":41,
+    "332":41,
+    "333":41,
+    "334":41,
+    "335":41,
+    "336":42,
+    "337":42,
+    "338":42,
+    "339":42,
+    "340":42,
+    "341":42,
+    "342":42,
+    "343":42,
+    "344":43,
+    "345":43,
+    "346":43,
+    "347":43,
+    "348":43,
+    "349":43,
+    "350":43,
+    "351":43,
+    "352":44,
+    "353":44,
+    "354":44,
+    "355":44,
+    "356":44,
+    "357":44,
+    "358":44,
+    "359":44,
+    "360":45,
+    "361":45,
+    "362":45,
+    "363":45,
+    "364":45,
+    "365":45,
+    "366":45,
+    "367":45,
+    "368":46,
+    "369":46,
+    "370":46,
+    "371":46,
+    "372":46,
+    "373":46,
+    "374":46,
+    "375":46,
+    "376":47,
+    "377":47,
+    "378":47,
+    "379":47,
+    "380":47,
+    "381":47,
+    "382":47,
+    "383":47,
+    "384":48,
+    "385":48,
+    "386":48,
+    "387":48,
+    "388":48,
+    "389":48,
+    "390":48,
+    "391":48,
+    "392":49,
+    "393":49,
+    "394":49,
+    "395":49,
+    "396":49,
+    "397":49,
+    "398":49,
+    "399":49,
+    "400":50,
+    "401":50,
+    "402":50,
+    "403":50,
+    "404":50,
+    "405":50,
+    "406":50,
+    "407":50,
+    "408":51,
+    "409":51,
+    "410":51,
+    "411":51,
+    "412":51,
+    "413":51,
+    "414":51,
+    "415":51,
+    "416":52,
+    "417":52,
+    "418":52,
+    "419":52,
+    "420":52,
+    "421":52,
+    "422":52,
+    "423":52,
+    "424":53,
+    "425":53,
+    "426":53,
+    "427":53,
+    "428":53,
+    "429":53,
+    "430":53,
+    "431":53,
+    "432":54,
+    "433":54,
+    "434":54,
+    "435":54,
+    "436":54,
+    "437":54,
+    "438":54,
+    "439":54,
+    "440":55,
+    "441":55,
+    "442":55,
+    "443":55,
+    "444":55,
+    "445":55,
+    "446":55,
+    "447":55,
+    "448":56,
+    "449":56,
+    "450":56,
+    "451":56,
+    "452":56,
+    "453":56,
+    "454":56,
+    "455":56,
+    "456":57,
+    "457":57,
+    "458":57,
+    "459":57,
+    "460":57,
+    "461":57,
+    "462":57,
+    "463":57,
+    "464":58,
+    "465":58,
+    "466":58,
+    "467":58,
+    "468":58,
+    "469":58,
+    "470":58,
+    "471":58,
+    "472":59,
+    "473":59,
+    "474":59,
+    "475":59,
+    "476":59,
+    "477":59,
+    "478":59,
+    "479":59,
+    "480":60,
+    "481":60,
+    "482":60,
+    "483":60,
+    "484":60,
+    "485":60,
+    "486":60,
+    "487":60,
+    "488":61,
+    "489":61,
+    "490":61,
+    "491":61,
+    "492":61,
+    "493":61,
+    "494":61,
+    "495":61,
+    "496":62,
+    "497":62,
+    "498":62,
+    "499":62,
+    "500":62,
+    "501":62,
+    "502":62,
+    "503":62,
+    "504":63,
+    "505":63,
+    "506":63,
+    "507":63,
+    "508":63,
+    "509":63,
+    "510":63,
+    "511":63,
+    "512":64,
+    "513":64,
+    "514":64,
+    "515":64,
+    "516":64,
+    "517":64,
+    "518":64,
+    "519":64,
+    "520":65,
+    "521":65,
+    "522":65,
+    "523":65,
+    "524":65,
+    "525":65,
+    "526":65,
+    "527":65,
+    "528":66,
+    "529":66,
+    "530":66,
+    "531":66,
+    "532":66,
+    "533":66,
+    "534":66,
+    "535":66,
+    "536":67,
+    "537":67,
+    "538":67,
+    "539":67,
+    "540":67,
+    "541":67,
+    "542":67,
+    "543":67,
+    "544":68,
+    "545":68,
+    "546":68,
+    "547":68,
+    "548":68,
+    "549":68,
+    "550":68,
+    "551":68,
+    "552":69,
+    "553":69,
+    "554":69,
+    "555":69,
+    "556":69,
+    "557":69,
+    "558":69,
+    "559":69,
+    "560":70,
+    "561":70,
+    "562":70,
+    "563":70,
+    "564":70,
+    "565":70,
+    "566":70,
+    "567":70,
+    "568":71,
+    "569":71,
+    "570":71,
+    "571":71,
+    "572":71,
+    "573":71,
+    "574":71,
+    "575":71,
+    "576":72,
+    "577":72,
+    "578":72,
+    "579":72,
+    "580":72,
+    "581":72,
+    "582":72,
+    "583":72,
+    "584":73,
+    "585":73,
+    "586":73,
+    "587":73,
+    "588":73,
+    "589":73,
+    "590":73,
+    "591":73,
+    "592":74,
+    "593":74,
+    "594":74,
+    "595":74,
+    "596":74,
+    "597":74,
+    "598":74,
+    "599":74,
+    "600":75,
+    "601":75,
+    "602":75,
+    "603":75,
+    "604":75,
+    "605":75,
+    "606":75,
+    "607":75,
+    "608":76,
+    "609":76,
+    "610":76,
+    "611":76,
+    "612":76,
+    "613":76,
+    "614":76,
+    "615":76,
+    "616":77,
+    "617":77,
+    "618":77,
+    "619":77,
+    "620":77,
+    "621":77,
+    "622":77,
+    "623":77,
+    "624":78,
+    "625":78,
+    "626":78,
+    "627":78,
+    "628":78,
+    "629":78,
+    "630":78,
+    "631":78,
+    "632":79,
+    "633":79,
+    "634":79,
+    "635":79,
+    "636":79,
+    "637":79,
+    "638":79,
+    "639":79,
+    "640":80,
+    "641":80,
+    "642":80,
+    "643":80,
+    "644":80,
+    "645":80,
+    "646":80,
+    "647":80,
+    "648":81,
+    "649":81,
+    "650":81,
+    "651":81,
+    "652":81,
+    "653":81,
+    "654":81,
+    "655":81,
+    "656":82,
+    "657":82,
+    "658":82,
+    "659":82,
+    "660":82,
+    "661":82,
+    "662":82,
+    "663":82,
+    "664":83,
+    "665":83,
+    "666":83,
+    "667":83,
+    "668":83,
+    "669":83,
+    "670":83,
+    "671":83,
+    "672":84,
+    "673":84,
+    "674":84,
+    "675":84,
+    "676":84,
+    "677":84,
+    "678":84,
+    "679":84,
+    "680":85,
+    "681":85,
+    "682":85,
+    "683":85,
+    "684":85,
+    "685":85,
+    "686":85,
+    "687":85,
+    "688":86,
+    "689":86,
+    "690":86,
+    "691":86,
+    "692":86,
+    "693":86,
+    "694":86,
+    "695":86,
+    "696":87,
+    "697":87,
+    "698":87,
+    "699":87,
+    "700":87,
+    "701":87,
+    "702":87,
+    "703":87,
+    "704":88,
+    "705":88,
+    "706":88,
+    "707":88,
+    "708":88,
+    "709":88,
+    "710":88,
+    "711":88,
+    "712":89,
+    "713":89,
+    "714":89,
+    "715":89,
+    "716":89,
+    "717":89,
+    "718":89,
+    "719":89,
+    "720":90,
+    "721":90,
+    "722":90,
+    "723":90,
+    "724":90,
+    "725":90,
+    "726":90,
+    "727":90,
+    "728":91,
+    "729":91,
+    "730":91,
+    "731":91,
+    "732":91,
+    "733":91,
+    "734":91,
+    "735":91,
+    "736":92,
+    "737":92,
+    "738":92,
+    "739":92,
+    "740":92,
+    "741":92,
+    "742":92,
+    "743":92,
+    "744":93,
+    "745":93,
+    "746":93,
+    "747":93,
+    "748":93,
+    "749":93,
+    "750":93,
+    "751":93,
+    "752":94,
+    "753":94,
+    "754":94,
+    "755":94,
+    "756":94,
+    "757":94,
+    "758":94,
+    "759":94,
+    "760":95,
+    "761":95,
+    "762":95,
+    "763":95,
+    "764":95,
+    "765":95,
+    "766":95,
+    "767":95,
+    "768":96,
+    "769":96,
+    "770":96,
+    "771":96,
+    "772":96,
+    "773":96,
+    "774":96,
+    "775":96,
+    "776":97,
+    "777":97,
+    "778":97,
+    "779":97,
+    "780":97,
+    "781":97,
+    "782":97,
+    "783":97,
+    "784":98,
+    "785":98,
+    "786":98,
+    "787":98,
+    "788":98,
+    "789":98,
+    "790":98,
+    "791":98,
+    "792":99,
+    "793":99,
+    "794":99,
+    "795":99,
+    "796":99,
+    "797":99,
+    "798":99,
+    "799":99,
+    "800":100,
+    "801":100,
+    "802":100,
+    "803":100,
+    "804":100,
+    "805":100,
+    "806":100,
+    "807":100,
+    "808":101,
+    "809":101,
+    "810":101,
+    "811":101,
+    "812":101,
+    "813":101,
+    "814":101,
+    "815":101,
+    "816":102,
+    "817":102,
+    "818":102,
+    "819":102,
+    "820":102,
+    "821":102,
+    "822":102,
+    "823":102,
+    "824":103,
+    "825":103,
+    "826":103,
+    "827":103,
+    "828":103,
+    "829":103,
+    "830":103,
+    "831":103,
+    "832":104,
+    "833":104,
+    "834":104,
+    "835":104,
+    "836":104,
+    "837":104,
+    "838":104,
+    "839":104,
+    "840":105,
+    "841":105,
+    "842":105,
+    "843":105,
+    "844":105,
+    "845":105,
+    "846":105,
+    "847":105,
+    "848":106,
+    "849":106,
+    "850":106,
+    "851":106,
+    "852":106,
+    "853":106,
+    "854":106,
+    "855":106,
+    "856":107,
+    "857":107,
+    "858":107,
+    "859":107,
+    "860":107,
+    "861":107,
+    "862":107,
+    "863":107,
+    "864":108,
+    "865":108,
+    "866":108,
+    "867":108,
+    "868":108,
+    "869":108,
+    "870":108,
+    "871":108,
+    "872":109,
+    "873":109,
+    "874":109,
+    "875":109,
+    "876":109,
+    "877":109,
+    "878":109,
+    "879":109,
+    "880":110,
+    "881":110,
+    "882":110,
+    "883":110,
+    "884":110,
+    "885":110,
+    "886":110,
+    "887":110,
+    "888":111,
+    "889":111,
+    "890":111,
+    "891":111,
+    "892":111,
+    "893":111,
+    "894":111,
+    "895":111,
+    "896":112,
+    "897":112,
+    "898":112,
+    "899":112,
+    "900":112,
+    "901":112,
+    "902":112,
+    "903":112,
+    "904":113,
+    "905":113,
+    "906":113,
+    "907":113,
+    "908":113,
+    "909":113,
+    "910":113,
+    "911":113,
+    "912":114,
+    "913":114,
+    "914":114,
+    "915":114,
+    "916":114,
+    "917":114,
+    "918":114,
+    "919":114,
+    "920":115,
+    "921":115,
+    "922":115,
+    "923":115,
+    "924":115,
+    "925":115,
+    "926":115,
+    "927":115,
+    "928":116,
+    "929":116,
+    "930":116,
+    "931":116,
+    "932":116,
+    "933":116,
+    "934":116,
+    "935":116,
+    "936":117,
+    "937":117,
+    "938":117,
+    "939":117,
+    "940":117,
+    "941":117,
+    "942":117,
+    "943":117,
+    "944":118,
+    "945":118,
+    "946":118,
+    "947":118,
+    "948":118,
+    "949":118,
+    "950":118,
+    "951":118,
+    "952":119,
+    "953":119,
+    "954":119,
+    "955":119,
+    "956":119,
+    "957":119,
+    "958":119,
+    "959":119,
+    "960":120,
+    "961":120,
+    "962":120,
+    "963":120,
+    "964":120,
+    "965":120,
+    "966":120,
+    "967":120,
+    "968":121,
+    "969":121,
+    "970":121,
+    "971":121,
+    "972":121,
+    "973":121,
+    "974":121,
+    "975":121,
+    "976":122,
+    "977":122,
+    "978":122,
+    "979":122,
+    "980":122,
+    "981":122,
+    "982":122,
+    "983":122,
+    "984":123,
+    "985":123,
+    "986":123,
+    "987":123,
+    "988":123,
+    "989":123,
+    "990":123,
+    "991":123,
+    "992":124,
+    "993":124,
+    "994":124,
+    "995":124,
+    "996":124,
+    "997":124,
+    "998":124,
+    "999":124
+}

src/hand_model_outputsize.py

@@ -0,0 +1,17 @@
+import torch
+from tqdm import tqdm
+import json
+
+from src.model import handpose_model
+
+model = handpose_model()
+
+size = {}
+for i in tqdm(range(10, 1000)):
+    data = torch.randn(1, 3, i, i)
+    if torch.cuda.is_available():
+        data = data.cuda()
+    size[i] = model(data).size(2)
+
+with open('hand_model_output_size.json') as f:
+    json.dump(size, f)

src/model.py

@@ -0,0 +1,219 @@
+import torch
+from collections import OrderedDict
+
+import torch
+import torch.nn as nn
+
+def make_layers(block, no_relu_layers):
+    layers = []
+    for layer_name, v in block.items():
+        if 'pool' in layer_name:
+            layer = nn.MaxPool2d(kernel_size=v[0], stride=v[1],
+                                    padding=v[2])
+            layers.append((layer_name, layer))
+        else:
+            conv2d = nn.Conv2d(in_channels=v[0], out_channels=v[1],
+                               kernel_size=v[2], stride=v[3],
+                               padding=v[4])
+            layers.append((layer_name, conv2d))
+            if layer_name not in no_relu_layers:
+                layers.append(('relu_'+layer_name, nn.ReLU(inplace=True)))
+
+    return nn.Sequential(OrderedDict(layers))
+
+class bodypose_model(nn.Module):
+    def __init__(self):
+        super(bodypose_model, self).__init__()
+
+        # these layers have no relu layer
+        no_relu_layers = ['conv5_5_CPM_L1', 'conv5_5_CPM_L2', 'Mconv7_stage2_L1',\
+                          'Mconv7_stage2_L2', 'Mconv7_stage3_L1', 'Mconv7_stage3_L2',\
+                          'Mconv7_stage4_L1', 'Mconv7_stage4_L2', 'Mconv7_stage5_L1',\
+                          'Mconv7_stage5_L2', 'Mconv7_stage6_L1', 'Mconv7_stage6_L1']
+        blocks = {}
+        block0 = OrderedDict([
+                      ('conv1_1', [3, 64, 3, 1, 1]),
+                      ('conv1_2', [64, 64, 3, 1, 1]),
+                      ('pool1_stage1', [2, 2, 0]),
+                      ('conv2_1', [64, 128, 3, 1, 1]),
+                      ('conv2_2', [128, 128, 3, 1, 1]),
+                      ('pool2_stage1', [2, 2, 0]),
+                      ('conv3_1', [128, 256, 3, 1, 1]),
+                      ('conv3_2', [256, 256, 3, 1, 1]),
+                      ('conv3_3', [256, 256, 3, 1, 1]),
+                      ('conv3_4', [256, 256, 3, 1, 1]),
+                      ('pool3_stage1', [2, 2, 0]),
+                      ('conv4_1', [256, 512, 3, 1, 1]),
+                      ('conv4_2', [512, 512, 3, 1, 1]),
+                      ('conv4_3_CPM', [512, 256, 3, 1, 1]),
+                      ('conv4_4_CPM', [256, 128, 3, 1, 1])
+                  ])
+
+
+        # Stage 1
+        block1_1 = OrderedDict([
+                        ('conv5_1_CPM_L1', [128, 128, 3, 1, 1]),
+                        ('conv5_2_CPM_L1', [128, 128, 3, 1, 1]),
+                        ('conv5_3_CPM_L1', [128, 128, 3, 1, 1]),
+                        ('conv5_4_CPM_L1', [128, 512, 1, 1, 0]),
+                        ('conv5_5_CPM_L1', [512, 38, 1, 1, 0])
+                    ])
+
+        block1_2 = OrderedDict([
+                        ('conv5_1_CPM_L2', [128, 128, 3, 1, 1]),
+                        ('conv5_2_CPM_L2', [128, 128, 3, 1, 1]),
+                        ('conv5_3_CPM_L2', [128, 128, 3, 1, 1]),
+                        ('conv5_4_CPM_L2', [128, 512, 1, 1, 0]),
+                        ('conv5_5_CPM_L2', [512, 19, 1, 1, 0])
+                    ])
+        blocks['block1_1'] = block1_1
+        blocks['block1_2'] = block1_2
+
+        self.model0 = make_layers(block0, no_relu_layers)
+
+        # Stages 2 - 6
+        for i in range(2, 7):
+            blocks['block%d_1' % i] = OrderedDict([
+                    ('Mconv1_stage%d_L1' % i, [185, 128, 7, 1, 3]),
+                    ('Mconv2_stage%d_L1' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv3_stage%d_L1' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv4_stage%d_L1' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv5_stage%d_L1' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv6_stage%d_L1' % i, [128, 128, 1, 1, 0]),
+                    ('Mconv7_stage%d_L1' % i, [128, 38, 1, 1, 0])
+                ])
+
+            blocks['block%d_2' % i] = OrderedDict([
+                    ('Mconv1_stage%d_L2' % i, [185, 128, 7, 1, 3]),
+                    ('Mconv2_stage%d_L2' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv3_stage%d_L2' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv4_stage%d_L2' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv5_stage%d_L2' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv6_stage%d_L2' % i, [128, 128, 1, 1, 0]),
+                    ('Mconv7_stage%d_L2' % i, [128, 19, 1, 1, 0])
+                ])
+
+        for k in blocks.keys():
+            blocks[k] = make_layers(blocks[k], no_relu_layers)
+
+        self.model1_1 = blocks['block1_1']
+        self.model2_1 = blocks['block2_1']
+        self.model3_1 = blocks['block3_1']
+        self.model4_1 = blocks['block4_1']
+        self.model5_1 = blocks['block5_1']
+        self.model6_1 = blocks['block6_1']
+
+        self.model1_2 = blocks['block1_2']
+        self.model2_2 = blocks['block2_2']
+        self.model3_2 = blocks['block3_2']
+        self.model4_2 = blocks['block4_2']
+        self.model5_2 = blocks['block5_2']
+        self.model6_2 = blocks['block6_2']
+
+
+    def forward(self, x):
+
+        out1 = self.model0(x)
+
+        out1_1 = self.model1_1(out1)
+        out1_2 = self.model1_2(out1)
+        out2 = torch.cat([out1_1, out1_2, out1], 1)
+
+        out2_1 = self.model2_1(out2)
+        out2_2 = self.model2_2(out2)
+        out3 = torch.cat([out2_1, out2_2, out1], 1)
+
+        out3_1 = self.model3_1(out3)
+        out3_2 = self.model3_2(out3)
+        out4 = torch.cat([out3_1, out3_2, out1], 1)
+
+        out4_1 = self.model4_1(out4)
+        out4_2 = self.model4_2(out4)
+        out5 = torch.cat([out4_1, out4_2, out1], 1)
+
+        out5_1 = self.model5_1(out5)
+        out5_2 = self.model5_2(out5)
+        out6 = torch.cat([out5_1, out5_2, out1], 1)
+
+        out6_1 = self.model6_1(out6)
+        out6_2 = self.model6_2(out6)
+
+        return out6_1, out6_2
+
+class handpose_model(nn.Module):
+    def __init__(self):
+        super(handpose_model, self).__init__()
+
+        # these layers have no relu layer
+        no_relu_layers = ['conv6_2_CPM', 'Mconv7_stage2', 'Mconv7_stage3',\
+                          'Mconv7_stage4', 'Mconv7_stage5', 'Mconv7_stage6']
+        # stage 1
+        block1_0 = OrderedDict([
+                ('conv1_1', [3, 64, 3, 1, 1]),
+                ('conv1_2', [64, 64, 3, 1, 1]),
+                ('pool1_stage1', [2, 2, 0]),
+                ('conv2_1', [64, 128, 3, 1, 1]),
+                ('conv2_2', [128, 128, 3, 1, 1]),
+                ('pool2_stage1', [2, 2, 0]),
+                ('conv3_1', [128, 256, 3, 1, 1]),
+                ('conv3_2', [256, 256, 3, 1, 1]),
+                ('conv3_3', [256, 256, 3, 1, 1]),
+                ('conv3_4', [256, 256, 3, 1, 1]),
+                ('pool3_stage1', [2, 2, 0]),
+                ('conv4_1', [256, 512, 3, 1, 1]),
+                ('conv4_2', [512, 512, 3, 1, 1]),
+                ('conv4_3', [512, 512, 3, 1, 1]),
+                ('conv4_4', [512, 512, 3, 1, 1]),
+                ('conv5_1', [512, 512, 3, 1, 1]),
+                ('conv5_2', [512, 512, 3, 1, 1]),
+                ('conv5_3_CPM', [512, 128, 3, 1, 1])
+            ])
+
+        block1_1 = OrderedDict([
+            ('conv6_1_CPM', [128, 512, 1, 1, 0]),
+            ('conv6_2_CPM', [512, 22, 1, 1, 0])
+        ])
+
+        blocks = {}
+        blocks['block1_0'] = block1_0
+        blocks['block1_1'] = block1_1
+
+        # stage 2-6
+        for i in range(2, 7):
+            blocks['block%d' % i] = OrderedDict([
+                    ('Mconv1_stage%d' % i, [150, 128, 7, 1, 3]),
+                    ('Mconv2_stage%d' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv3_stage%d' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv4_stage%d' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv5_stage%d' % i, [128, 128, 7, 1, 3]),
+                    ('Mconv6_stage%d' % i, [128, 128, 1, 1, 0]),
+                    ('Mconv7_stage%d' % i, [128, 22, 1, 1, 0])
+                ])
+
+        for k in blocks.keys():
+            blocks[k] = make_layers(blocks[k], no_relu_layers)
+
+        self.model1_0 = blocks['block1_0']
+        self.model1_1 = blocks['block1_1']
+        self.model2 = blocks['block2']
+        self.model3 = blocks['block3']
+        self.model4 = blocks['block4']
+        self.model5 = blocks['block5']
+        self.model6 = blocks['block6']
+
+    def forward(self, x):
+        out1_0 = self.model1_0(x)
+        out1_1 = self.model1_1(out1_0)
+        concat_stage2 = torch.cat([out1_1, out1_0], 1)
+        out_stage2 = self.model2(concat_stage2)
+        concat_stage3 = torch.cat([out_stage2, out1_0], 1)
+        out_stage3 = self.model3(concat_stage3)
+        concat_stage4 = torch.cat([out_stage3, out1_0], 1)
+        out_stage4 = self.model4(concat_stage4)
+        concat_stage5 = torch.cat([out_stage4, out1_0], 1)
+        out_stage5 = self.model5(concat_stage5)
+        concat_stage6 = torch.cat([out_stage5, out1_0], 1)
+        out_stage6 = self.model6(concat_stage6)
+        return out_stage6
+
+

src/util.py

@@ -0,0 +1,198 @@
+import numpy as np
+import math
+import cv2
+import matplotlib
+from matplotlib.backends.backend_agg import FigureCanvasAgg as FigureCanvas
+from matplotlib.figure import Figure
+import numpy as np
+import matplotlib.pyplot as plt
+import cv2
+
+
+def padRightDownCorner(img, stride, padValue):
+    h = img.shape[0]
+    w = img.shape[1]
+
+    pad = 4 * [None]
+    pad[0] = 0 # up
+    pad[1] = 0 # left
+    pad[2] = 0 if (h % stride == 0) else stride - (h % stride) # down
+    pad[3] = 0 if (w % stride == 0) else stride - (w % stride) # right
+
+    img_padded = img
+    pad_up = np.tile(img_padded[0:1, :, :]*0 + padValue, (pad[0], 1, 1))
+    img_padded = np.concatenate((pad_up, img_padded), axis=0)
+    pad_left = np.tile(img_padded[:, 0:1, :]*0 + padValue, (1, pad[1], 1))
+    img_padded = np.concatenate((pad_left, img_padded), axis=1)
+    pad_down = np.tile(img_padded[-2:-1, :, :]*0 + padValue, (pad[2], 1, 1))
+    img_padded = np.concatenate((img_padded, pad_down), axis=0)
+    pad_right = np.tile(img_padded[:, -2:-1, :]*0 + padValue, (1, pad[3], 1))
+    img_padded = np.concatenate((img_padded, pad_right), axis=1)
+
+    return img_padded, pad
+
+# transfer caffe model to pytorch which will match the layer name
+def transfer(model, model_weights):
+    transfered_model_weights = {}
+    for weights_name in model.state_dict().keys():
+        transfered_model_weights[weights_name] = model_weights['.'.join(weights_name.split('.')[1:])]
+    return transfered_model_weights
+
+# draw the body keypoint and lims
+def draw_bodypose(canvas, candidate, subset):
+    stickwidth = 4
+    limbSeq = [[2, 3], [2, 6], [3, 4], [4, 5], [6, 7], [7, 8], [2, 9], [9, 10], \
+               [10, 11], [2, 12], [12, 13], [13, 14], [2, 1], [1, 15], [15, 17], \
+               [1, 16], [16, 18], [3, 17], [6, 18]]
+
+    colors = [[255, 0, 0], [255, 85, 0], [255, 170, 0], [255, 255, 0], [170, 255, 0], [85, 255, 0], [0, 255, 0], \
+              [0, 255, 85], [0, 255, 170], [0, 255, 255], [0, 170, 255], [0, 85, 255], [0, 0, 255], [85, 0, 255], \
+              [170, 0, 255], [255, 0, 255], [255, 0, 170], [255, 0, 85]]
+    for i in range(18):
+        for n in range(len(subset)):
+            index = int(subset[n][i])
+            if index == -1:
+                continue
+            x, y = candidate[index][0:2]
+            cv2.circle(canvas, (int(x), int(y)), 4, colors[i], thickness=-1)
+    for i in range(17):
+        for n in range(len(subset)):
+            index = subset[n][np.array(limbSeq[i]) - 1]
+            if -1 in index:
+                continue
+            cur_canvas = canvas.copy()
+            Y = candidate[index.astype(int), 0]
+            X = candidate[index.astype(int), 1]
+            mX = np.mean(X)
+            mY = np.mean(Y)
+            length = ((X[0] - X[1]) ** 2 + (Y[0] - Y[1]) ** 2) ** 0.5
+            angle = math.degrees(math.atan2(X[0] - X[1], Y[0] - Y[1]))
+            polygon = cv2.ellipse2Poly((int(mY), int(mX)), (int(length / 2), stickwidth), int(angle), 0, 360, 1)
+            cv2.fillConvexPoly(cur_canvas, polygon, colors[i])
+            canvas = cv2.addWeighted(canvas, 0.4, cur_canvas, 0.6, 0)
+    # plt.imsave("preview.jpg", canvas[:, :, [2, 1, 0]])
+    # plt.imshow(canvas[:, :, [2, 1, 0]])
+    return canvas
+
+def draw_handpose(canvas, all_hand_peaks, show_number=False):
+    edges = [[0, 1], [1, 2], [2, 3], [3, 4], [0, 5], [5, 6], [6, 7], [7, 8], [0, 9], [9, 10], \
+             [10, 11], [11, 12], [0, 13], [13, 14], [14, 15], [15, 16], [0, 17], [17, 18], [18, 19], [19, 20]]
+    fig = Figure(figsize=plt.figaspect(canvas))
+
+    fig.subplots_adjust(0, 0, 1, 1)
+    fig.subplots_adjust(bottom=0, top=1, left=0, right=1)
+    bg = FigureCanvas(fig)
+    ax = fig.subplots()
+    ax.axis('off')
+    ax.imshow(canvas)
+
+    width, height = ax.figure.get_size_inches() * ax.figure.get_dpi()
+
+    for peaks in all_hand_peaks:
+        for ie, e in enumerate(edges):
+            if np.sum(np.all(peaks[e], axis=1)==0)==0:
+                x1, y1 = peaks[e[0]]
+                x2, y2 = peaks[e[1]]
+                ax.plot([x1, x2], [y1, y2], color=matplotlib.colors.hsv_to_rgb([ie/float(len(edges)), 1.0, 1.0]))
+
+        for i, keyponit in enumerate(peaks):
+            x, y = keyponit
+            ax.plot(x, y, 'r.')
+            if show_number:
+                ax.text(x, y, str(i))
+    bg.draw()
+    canvas = np.fromstring(bg.tostring_rgb(), dtype='uint8').reshape(int(height), int(width), 3)
+    return canvas
+
+# image drawed by opencv is not good.
+def draw_handpose_by_opencv(canvas, peaks, show_number=False):
+    edges = [[0, 1], [1, 2], [2, 3], [3, 4], [0, 5], [5, 6], [6, 7], [7, 8], [0, 9], [9, 10], \
+             [10, 11], [11, 12], [0, 13], [13, 14], [14, 15], [15, 16], [0, 17], [17, 18], [18, 19], [19, 20]]
+    # cv2.rectangle(canvas, (x, y), (x+w, y+w), (0, 255, 0), 2, lineType=cv2.LINE_AA)
+    # cv2.putText(canvas, 'left' if is_left else 'right', (x, y), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
+    for ie, e in enumerate(edges):
+        if np.sum(np.all(peaks[e], axis=1)==0)==0:
+            x1, y1 = peaks[e[0]]
+            x2, y2 = peaks[e[1]]
+            cv2.line(canvas, (x1, y1), (x2, y2), matplotlib.colors.hsv_to_rgb([ie/float(len(edges)), 1.0, 1.0])*255, thickness=2)
+
+    for i, keyponit in enumerate(peaks):
+        x, y = keyponit
+        cv2.circle(canvas, (x, y), 4, (0, 0, 255), thickness=-1)
+        if show_number:
+            cv2.putText(canvas, str(i), (x, y), cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 0), lineType=cv2.LINE_AA)
+    return canvas
+
+# detect hand according to body pose keypoints
+# please refer to https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/src/openpose/hand/handDetector.cpp
+def handDetect(candidate, subset, oriImg):
+    # right hand: wrist 4, elbow 3, shoulder 2
+    # left hand: wrist 7, elbow 6, shoulder 5
+    ratioWristElbow = 0.33
+    detect_result = []
+    image_height, image_width = oriImg.shape[0:2]
+    for person in subset.astype(int):
+        # if any of three not detected
+        has_left = np.sum(person[[5, 6, 7]] == -1) == 0
+        has_right = np.sum(person[[2, 3, 4]] == -1) == 0
+        if not (has_left or has_right):
+            continue
+        hands = []
+        #left hand
+        if has_left:
+            left_shoulder_index, left_elbow_index, left_wrist_index = person[[5, 6, 7]]
+            x1, y1 = candidate[left_shoulder_index][:2]
+            x2, y2 = candidate[left_elbow_index][:2]
+            x3, y3 = candidate[left_wrist_index][:2]
+            hands.append([x1, y1, x2, y2, x3, y3, True])
+        # right hand
+        if has_right:
+            right_shoulder_index, right_elbow_index, right_wrist_index = person[[2, 3, 4]]
+            x1, y1 = candidate[right_shoulder_index][:2]
+            x2, y2 = candidate[right_elbow_index][:2]
+            x3, y3 = candidate[right_wrist_index][:2]
+            hands.append([x1, y1, x2, y2, x3, y3, False])
+
+        for x1, y1, x2, y2, x3, y3, is_left in hands:
+            # pos_hand = pos_wrist + ratio * (pos_wrist - pos_elbox) = (1 + ratio) * pos_wrist - ratio * pos_elbox
+            # handRectangle.x = posePtr[wrist*3] + ratioWristElbow * (posePtr[wrist*3] - posePtr[elbow*3]);
+            # handRectangle.y = posePtr[wrist*3+1] + ratioWristElbow * (posePtr[wrist*3+1] - posePtr[elbow*3+1]);
+            # const auto distanceWristElbow = getDistance(poseKeypoints, person, wrist, elbow);
+            # const auto distanceElbowShoulder = getDistance(poseKeypoints, person, elbow, shoulder);
+            # handRectangle.width = 1.5f * fastMax(distanceWristElbow, 0.9f * distanceElbowShoulder);
+            x = x3 + ratioWristElbow * (x3 - x2)
+            y = y3 + ratioWristElbow * (y3 - y2)
+            distanceWristElbow = math.sqrt((x3 - x2) ** 2 + (y3 - y2) ** 2)
+            distanceElbowShoulder = math.sqrt((x2 - x1) ** 2 + (y2 - y1) ** 2)
+            width = 1.5 * max(distanceWristElbow, 0.9 * distanceElbowShoulder)
+            # x-y refers to the center --> offset to topLeft point
+            # handRectangle.x -= handRectangle.width / 2.f;
+            # handRectangle.y -= handRectangle.height / 2.f;
+            x -= width / 2
+            y -= width / 2  # width = height
+            # overflow the image
+            if x < 0: x = 0
+            if y < 0: y = 0
+            width1 = width
+            width2 = width
+            if x + width > image_width: width1 = image_width - x
+            if y + width > image_height: width2 = image_height - y
+            width = min(width1, width2)
+            # the max hand box value is 20 pixels
+            if width >= 20:
+                detect_result.append([int(x), int(y), int(width), is_left])
+
+    '''
+    return value: [[x, y, w, True if left hand else False]].
+    width=height since the network require squared input.
+    x, y is the coordinate of top left 
+    '''
+    return detect_result
+
+# get max index of 2d array
+def npmax(array):
+    arrayindex = array.argmax(1)
+    arrayvalue = array.max(1)
+    i = arrayvalue.argmax()
+    j = arrayindex[i]
+    return i, j