change model loading methods

app.py

@@ -14,6 +14,7 @@ img_examples = [["examples/group_test.png"],
                 ["examples/sad-1.png"],
                 ["examples/sad-2.png"]]
 
+ai_models = Classifier()
 
 def predict(video_in, image_in_video, image_in_img):
 
@@ -25,26 +26,25 @@ def predict(video_in, image_in_video, image_in_img):
         fname, fext = os.path.splitext(img_path)
 
         if fext == ".mp4":
-            classifier = Classifier(is_image=False, input_path=img_path)
+            ai_models.check_file_type(is_image=False, input_path=img_path)
         else:
-            classifier = Classifier(is_image=True, input_path=img_path)
+            ai_models.check_file_type(is_image=True, input_path=img_path)
 
-        output_path, pred_time = classifier.run()
+        output_path, pred_time = ai_models.run()
         gr.Markdown(f"Total prediction time: ", pred_time)
         return output_path, pred_time
 
     if video_in:
         fname, fext = os.path.splitext(video_in)
         if fext == ".mp4":
-            classifier = Classifier(is_image=False, input_path=video_in)
+            ai_models.check_file_type(is_image=False, input_path=video_in)
         else:
-            classifier = Classifier(is_image=True, input_path=video_in)
+            ai_models.check_file_type(is_image=True, input_path=video_in)
 
-        output_path, pred_time = classifier.run()
+        output_path, pred_time = ai_models.run()
         gr.Markdown(f"Total prediction time: ", pred_time)
         return output_path, pred_time
 
-
 def toggle(choice):
     if choice == "webcam":
         return gr.update(visible=True, value=None), gr.update(visible=False, value=None)
@@ -67,10 +67,9 @@ with gr.Blocks() as demo:
                 video_out = gr.Video(format='mp4')
 
         run_btn = gr.Button("Predict")
-        pred_tbox = gr.Textbox(label="Prediction Time")
-        run_btn.click(fn=predict, inputs=[video_in], outputs=[video_out, pred_tbox])
+        run_btn.click(fn=predict, inputs=[video_in], outputs=[video_out])
         gr.Examples(fn=predict, examples=video_examples, inputs=[
-                    video_in], outputs=[video_out, pred_tbox])
+                    video_in], outputs=[video_out])
 
     with gr.Tab("Image"):
         with gr.Row():

classifier.py

@@ -3,33 +3,38 @@ import os
 import numpy as np
 from tensorflow.keras.optimizers import Adam
 from tensorflow.keras.models import load_model
+from tensorflow.keras.models import model_from_json
 import time
 
 
 class Classifier:
-    def __init__(self, is_image, input_path):
-        self.is_image = is_image
-        self.counter = 0
-
-        self.classifier = load_model('model.h5')
-        self.face_detector = cv2.CascadeClassifier(
-            'haarcascade_frontalface_default.xml')
-        cv2.ocl.setUseOpenCL(False)
-
+    def __init__(self):
+        self.is_image = None
+        self.classifier = None
+        self.face_detector = None
         self.labels = {0: "Angry", 1: "Disgusted", 2: "Fearful",
                        3: "Happy", 4: "Neutral", 5: "Sad", 6: "Surprised"}
-
+        self.load_models()
+    
+    def check_file_type(self, is_image, input_path):
+        self.is_image = is_image
         self.input_path = input_path
-        self.output_path = "/tmp/gradio/result/" + self.input_path.split('/')[-1]
-        os.makedirs("/tmp/gradio/result", exist_ok=True)
-        is_dir = os.path.exists("/tmp/gradio/result")
-        print("Is there dir? :", is_dir)
-
-        print("Models have been loaded")
-    # def get_model(self):
-    #     model = load_model(model_path, compile=False)
-    #     model.compile(optimizer=Adam(learning_rate=1e-3),
-    #                 loss='categorical_crossentropy', metrics=['accuracy'])
+        self.output_path = "result/" + self.input_path.split('/')[-1]
+        os.makedirs("result", exist_ok=True)
+
+    def load_models(self):
+        # Load VGG
+        yaml_file = open('models/VGG19.yaml', 'r')
+        self.classifier = model_from_json(yaml_file.read())
+        self.classifier.load_weights("models/VGG19.h5")
+        yaml_file.close()
+        print("VGG19 has been loaded")
+
+        # Load Face Detector
+        self.face_detector = cv2.CascadeClassifier(
+        'models/haarcascade_frontalface_default.xml')
+        cv2.ocl.setUseOpenCL(False)
+        print("Face Detector has been loaded")
 
     def predict(self, frame):
 
@@ -40,12 +45,15 @@ class Classifier:
         for (x, y, w, h) in faces:
             cv2.rectangle(frame, (x, y-50), (x+w, y+h+10), (255, 0, 0), 2)
             roi_gray = gray[y:y + h, x:x + w]
+            roi_color = frame[y:y + h, x:x + w]
+            # Convert grayscale to RGB
+            roi_color_rgb = cv2.cvtColor(roi_color, cv2.COLOR_BGR2RGB)
             cropped_img = np.expand_dims(np.expand_dims(
-                cv2.resize(roi_gray, (48, 48)), -1), 0)
+                cv2.resize(roi_color_rgb, (48, 48)), 0), -1)
             prediction = self.classifier.predict(cropped_img)
             maxindex = int(np.argmax(prediction))
             cv2.putText(frame, self.labels[maxindex], (x+20, y-60),
-                        cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 0, 255), 2, cv2.LINE_AA)
+                        cv2.FONT_HERSHEY_SIMPLEX, 1.4, (0, 0, 255), 2, cv2.LINE_AA)
 
         return frame
 
@@ -64,7 +72,7 @@ class Classifier:
             fps = int(cap.get(5))
 
             # Define the codec and create a VideoWriter object to save the output video as .MOV
-            fourcc = cv2.VideoWriter_fourcc(*'MP4V')
+            fourcc = cv2.VideoWriter_fourcc('a', 'v', 'c', '1')
             out = cv2.VideoWriter(self.output_path, fourcc, fps,
                                   (frame_width, frame_height))
 

model.h5

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:8fb504950f1185882c2997ba0d9be2b62bbb85d05153827094a0f631819728cc
-size 12439824