Update app.py

app.py

@@ -64,21 +64,19 @@ def inference(video, lang, time_step):
     while success:
         if count % (int(frame_rate * time_step)) == 0:
             for i, box in enumerate(largest_boxes):
-                bbox_pos = box_position(box)
-                if np.linalg.norm(np.array(bbox_pos) - np.array(positions[i])) < 50:
-                    x1, y1 = box[0][0]
-                    x2, y2 = box[0][2]
-                    box_width = x2 - x1
-                    box_height = y2 - y1
-                    ratio = 0.2
-                    x1 = max(0, int(x1 - ratio * box_width))
-                    x2 = min(frame.shape[1], int(x2 + ratio * box_width))
-                    y1 = max(0, int(y1 - ratio * box_height))
-                    y2 = min(frame.shape[0], int(y2 + ratio * box_height))
-                    cropped_frame = frame[y1:y2, x1:x2]
-                    text = reader.readtext(cropped_frame)
-                    if text:
-                        temporal_profiles[i].append((count / frame_rate, text[0][1]))
+                x1, y1 = box[0][0]
+                x2, y2 = box[0][2]
+                box_width = x2 - x1
+                box_height = y2 - y1
+                ratio = 0.2
+                x1 = max(0, int(x1 - ratio * box_width))
+                x2 = min(frame.shape[1], int(x2 + ratio * box_width))
+                y1 = max(0, int(y1 - ratio * box_height))
+                y2 = min(frame.shape[0], int(y2 + ratio * box_height))
+                cropped_frame = frame[y1:y2, x1:x2]
+                text = reader.readtext(cropped_frame)
+                if text:
+                    temporal_profiles[i].append((count / frame_rate, text[0][1]))
             im = PIL.Image.fromarray(frame)
             im_with_boxes = draw_boxes(im, bounds)
             output_frames.append(np.array(im_with_boxes))