upgraded

app.py

@@ -68,6 +68,7 @@ class UI():
 
         with gr.Blocks() as demo:
             self.faces = gr.State([])
+            self.distance_type = gr.State("cosine")
             gr.Markdown("## FaceAnalyzer face recognition test")
             with gr.Tabs():
                 with gr.TabItem('Realtime Recognize'):
@@ -145,36 +146,16 @@ class UI():
                     self.cb_draw_landmarks.change(self.set_draw_landmarks, self.cb_draw_landmarks)
                     self.sld_nb_faces = gr.Slider(1,50,3,label="Maximum number of faces")
                     self.sld_nb_faces.change(self.set_nb_faces, self.sld_nb_faces)
+                    self.rd_distance_type = gr.Radio(
+                        ["cosine", "L1", "L2"], label="Distance", value="cosine"
+                    )
+                    self.rd_distance_type.change(self.change_distance, self.rd_distance_type, self.distance_type)
                     
 
         demo.queue().launch()
 
-    
-
-    def draw_name_on_bbox(self, image, bbox, name, font=cv2.FONT_HERSHEY_SIMPLEX,
-                        font_scale=1, thickness=2, color=(0, 0, 255)):
-        # Upscale the image to avoid pixelization of the text
-        height, width = image.shape[:2]
-        scale = max(1, int(max(height, width) / 800))
-        new_height, new_width = int(height * scale), int(width * scale)
-        resized_image = cv2.resize(image, (new_width, new_height))
-
-        # Upscale the bounding box
-        bbox = [int(val * scale) for val in bbox]
-
-        # Draw the bounding box and the name
-        x1, y1, x2, y2 = bbox
-        cv2.rectangle(resized_image, (x1, y1), (x2, y2), color, thickness)
-        text_size, _ = cv2.getTextSize(name, font, font_scale, thickness)
-        text_x = x1 + (x2 - x1 - text_size[0]) // 2
-        text_y = y1 - text_size[1] - 5
-        cv2.putText(resized_image, name, (text_x, text_y), font, font_scale, color, thickness)
-
-        # Downscale the image back to its original size
-        final_image = cv2.resize(resized_image, (width, height))
-
-        return final_image
-
+    def change_distance(self, type):
+        self.distance_type=type
 
     def clear_galery(self):
         return self.gallery.update(value=[]), self.add_file.update(value=[])
@@ -379,9 +360,9 @@ class UI():
 
         if fa.nb_faces>0:
             bboxes_and_names=[]
-            for i in range(fa.nb_faces):
+            for j in range(fa.nb_faces):
                 try:
-                    face = fa.faces[i]
+                    face = fa.faces[j]
                     vertices = face.get_face_outer_vertices()
                     face_image = face.getFaceBox(image, vertices, margins=(40,40,40,40))
                     embedding = DeepFace.represent(face_image, enforce_detection=False)[0]["embedding"]
@@ -390,13 +371,15 @@ class UI():
                     nearest_distance = 1e100
                     nearest = 0
                     for i, known_face in enumerate(self.known_faces):
-                        # absolute distance
-                        distance = np.abs(known_face["mean"]-embedding).sum()
-                        # euclidian distance
-                        #diff = known_face["mean"]-embedding
-                        #distance = np.sqrt([email protected])
-                        # Cosine distance
-                        distance = self.cosine_distance(known_face["mean"], embedding)
+                        if self.distance_type == "cosine":
+                            # Cosine distance
+                            distance = self.cosine_distance(known_face["mean"], embedding)
+                        elif self.distance_type =="L1":
+                            # absolute distance
+                            distance = np.abs(known_face["mean"]-embedding).sum()
+                        elif self.distance_type == "L2":
+                            # absolute distance
+                            distance = np.sqrt(np.square(known_face["mean"]-embedding).sum())
                         if distance<nearest_distance:
                             nearest_distance = distance
                             nearest = i
@@ -419,9 +402,9 @@ class UI():
         fa.process(image)
 
         if fa.nb_faces>0:
-            for i in range(fa.nb_faces):
+            for j in range(fa.nb_faces):
                 try:
-                    face = fa.faces[i]
+                    face = fa.faces[j]
                     vertices = face.get_face_outer_vertices()
                     face_image = face.getFaceBox(image, vertices, margins=(40,40,40,40))
                     embedding = DeepFace.represent(face_image, enforce_detection=False)[0]["embedding"]
@@ -430,13 +413,15 @@ class UI():
                     nearest_distance = 1e100
                     nearest = 0
                     for i, known_face in enumerate(self.known_faces):
-                        # absolute distance
-                        distance = np.abs(known_face["mean"]-embedding).sum()
-                        # euclidian distance
-                        #diff = known_face["mean"]-embedding
-                        #distance = np.sqrt([email protected])
-                        # Cosine distance
-                        distance = self.cosine_distance(known_face["mean"], embedding)
+                        if self.distance_type == "cosine":
+                            # Cosine distance
+                            distance = self.cosine_distance(known_face["mean"], embedding)
+                        elif self.distance_type =="L1":
+                            # absolute distance
+                            distance = np.abs(known_face["mean"]-embedding).sum()
+                        elif self.distance_type == "L2":
+                            # absolute distance
+                            distance = np.sqrt(np.square(known_face["mean"]-embedding).sum())
                         if distance<nearest_distance:
                             nearest_distance = distance
                             nearest = i