1

app.py

@@ -1,35 +1,111 @@
 import gradio as gr
-from transformers import SegformerFeatureExtractor, SegformerForSemanticSegmentation
-from PIL import Image
+
+from matplotlib import gridspec
+import matplotlib.pyplot as plt
 import numpy as np
+from PIL import Image
+import tensorflow as tf
+from transformers import SegformerFeatureExtractor, TFSegformerForSemanticSegmentation
+
+feature_extractor = SegformerFeatureExtractor.from_pretrained(
+    "nvidia/segformer-b5-finetuned-cityscapes-1024-1024"
+)
+model = TFSegformerForSemanticSegmentation.from_pretrained(
+    "nvidia/segformer-b5-finetuned-cityscapes-1024-1024"
+)
+
+def ade_palette():
+    """ADE20K palette that maps each class to RGB values."""
+    return [
+        [255,0,0], #빨강
+        [255,228,0], #노랑
+        [171,242,0], # 연두
+        [0,216,255], #하늘
+        [0,0,255], #파랑
+        [255,0,221], #핑크
+        [116,116,116], #회색
+        [95,0,255], #보라
+        [255,94,0], #주황
+        [71,200,62], #초록
+        [153,0,76], #마젠타
+        [67,116,217], #애매한하늘 + 파랑
+        [153,112,0], #겨자
+        [87,129,0], #녹색
+        [255,169,169], #분홍분홍
+        [35,30,183], #어두운 파랑
+        [225,186,133], #살색
+        [206,251,201] #연한초록
+        [165,102,255] #애매한 보라
+    ]
+
+labels_list = []
+
+with open(r'labels.txt', 'r') as fp:
+    for line in fp:
+        labels_list.append(line[:-1])
+
+colormap = np.asarray(ade_palette())
+
+def label_to_color_image(label):
+    if label.ndim != 2:
+        raise ValueError("Expect 2-D input label")
 
-# 모델과 특징 추출기 불러오기
-feature_extractor = SegformerFeatureExtractor.from_pretrained("nvidia/segformer-b5-finetuned-cityscapes-1024-1024")
-model = SegformerForSemanticSegmentation.from_pretrained("nvidia/segformer-b5-finetuned-cityscapes-1024-1024")
+    if np.max(label) >= len(colormap):
+        raise ValueError("label value too large.")
+    return colormap[label]
 
-def segment_image(image):
-    # 이미지를 처리하고 모델에 전달하기
-    inputs = feature_extractor(images=image, return_tensors="pt")
+def draw_plot(pred_img, seg):
+    fig = plt.figure(figsize=(20, 15))
+
+    grid_spec = gridspec.GridSpec(1, 2, width_ratios=[6, 1])
+
+    plt.subplot(grid_spec[0])
+    plt.imshow(pred_img)
+    plt.axis('off')
+    LABEL_NAMES = np.asarray(labels_list)
+    FULL_LABEL_MAP = np.arange(len(LABEL_NAMES)).reshape(len(LABEL_NAMES), 1)
+    FULL_COLOR_MAP = label_to_color_image(FULL_LABEL_MAP)
+
+    unique_labels = np.unique(seg.numpy().astype("uint8"))
+    ax = plt.subplot(grid_spec[1])
+    plt.imshow(FULL_COLOR_MAP[unique_labels].astype(np.uint8), interpolation="nearest")
+    ax.yaxis.tick_right()
+    plt.yticks(range(len(unique_labels)), LABEL_NAMES[unique_labels])
+    plt.xticks([], [])
+    ax.tick_params(width=0.0, labelsize=25)
+    return fig
+
+def sepia(input_img):
+    input_img = Image.fromarray(input_img)
+
+    inputs = feature_extractor(images=input_img, return_tensors="tf")
     outputs = model(**inputs)
     logits = outputs.logits
 
-    # 결과 처리 및 이미지로 변환
-    result = logits.argmax(dim=1)[0]
-    result = result.cpu().detach().numpy()
-    result_image = Image.fromarray(result.astype(np.uint8), mode="P")
-
-    # 결과 이미지 반환
-    return result_image
-
-# Gradio 인터페이스 정의
-iface = gr.Interface(
-    fn=segment_image,
-    inputs=gr.inputs.Image(type="pil"),
-    examples = ['image1.jpg', 'image2.jpg', 'image3.jpg'],
-    outputs="image",
-    title="SegFormer Image Segmentation",
-    description="Upload an image to segment it using the SegFormer model trained on Cityscapes dataset."
-)
+    logits = tf.transpose(logits, [0, 2, 3, 1])
+    logits = tf.image.resize(
+        logits, input_img.size[::-1]
+    )  # We reverse the shape of `image` because `image.size` returns width and height.
+    seg = tf.math.argmax(logits, axis=-1)[0]
+
+    color_seg = np.zeros(
+        (seg.shape[0], seg.shape[1], 3), dtype=np.uint8
+    )  # height, width, 3
+    for label, color in enumerate(colormap):
+        color_seg[seg.numpy() == label, :] = color
+
+    # Show image + mask
+    pred_img = np.array(input_img) * 0.5 + color_seg * 0.5
+    pred_img = pred_img.astype(np.uint8)
+
+    fig = draw_plot(pred_img, seg)
+    return fig
+
+demo = gr.Interface(fn=sepia,
+                    inputs=gr.Image(shape=(400, 600)),
+                    outputs=['plot'],
+                    examples=["image1", "image2", "image3"],
+                    allow_flagging='never')
+
 
-# 인터페이스 실행
-iface.launch()
+demo.launch()