changed sampling algorithem

app.py

@@ -40,30 +40,21 @@ def download_video(url,format_id):
     save_location = meta['id'] + '.' + meta['ext']
     return(save_location)
 
-def read_frames(dest_path):
-  original_images = []
-  images = []
-  for filename in sorted(dest_path.glob('*.jpg'),key=lambda p: int(p.stem)):
-    image = Image.open(filename).convert("RGB")
-    original_images.append(image)
-    images.append(preprocess(image))
-  return original_images, images
-
 def process_video_parallel(video, skip_frames, dest_path, num_processes, process_number):
     cap = cv2.VideoCapture(video)
-    chunks_per_process = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) // (num_processes * skip_frames)
-    count =  skip_frames * chunks_per_process * process_number
-    print(f"worker: {process_number}, process frames {count} ~ {skip_frames * chunks_per_process * (process_number + 1)} \n total number of frames: {cap.get(cv2.CAP_PROP_FRAME_COUNT)} \n video: {video}; isOpen? : {cap.isOpened()}")
-    while count < skip_frames * chunks_per_process * (process_number + 1) :
-        if skip_frames > 1: 
-            cap.set(cv2.CAP_PROP_POS_FRAMES, count)
+    frames_per_process = int(cap.get(cv2.CAP_PROP_FRAME_COUNT)) // (num_processes)
+    count =  frames_per_process * process_number
+    print(f"worker: {process_number}, process frames {count} ~ {frames_per_process * (process_number + 1)} \n total number of frames: {cap.get(cv2.CAP_PROP_FRAME_COUNT)} \n video: {video}; isOpen? : {cap.isOpened()}")
+    while count < frames_per_process * (process_number + 1) :
         ret, frame = cap.read()
         if not ret:
             break
-        filename =f"{dest_path}/{count}.jpg"
-        cv2.imwrite(filename, frame)
-        print(f"saved {filename}")
-        count += skip_frames  # Skip 300 frames i.e. 10 seconds for 30 fps
+        count += 1
+        if (count - frames_per_process * process_number) % skip_frames ==0:
+          filename =f"{dest_path}/{count}.jpg"
+          cv2.imwrite(filename, frame)
+          #print(f"saved {filename}")
+        
     cap.release()
 
 
@@ -87,8 +78,7 @@ def vid2frames(url, sampling_interval=1, ext='mp4'):
     except:
         skip_frames = int(30 * sampling_interval)
 
-    # testing
-    skip_frames = 1
+
     print(f'video saved at: {video}, fps:{fps}, skip_frames: {skip_frames}')
     # extract video frames at given sampling interval with multiprocessing - 
     print('extracting frames...')
@@ -99,7 +89,16 @@ def vid2frames(url, sampling_interval=1, ext='mp4'):
     print(f'n_workers: {n_workers}')
     with Pool(n_workers) as pool:
         pool.map(partial(process_video_parallel, video, skip_frames, dest_path, n_workers), range(n_workers))
-    return dest_path
+    # read frames
+    original_images = []
+    images = []
+    filenames = sorted(dest_path.glob('*.jpg'),key=lambda p: int(p.stem))
+    print(f"extracted {len(filenames)} frames")
+    for filename in filenames:
+      image = Image.open(filename).convert("RGB")
+      original_images.append(image)
+      images.append(preprocess(image))
+    return original_images, images
 
 
 def captioned_strip(images, caption=None, times=None, rows=1):
@@ -126,8 +125,7 @@ def captioned_strip(images, caption=None, times=None, rows=1):
     return img
 
 def run_inference(url, sampling_interval, search_query):
-    path_frames = vid2frames(url,sampling_interval)
-    original_images, images = read_frames(path_frames)
+    original_images, images = vid2frames(url,sampling_interval)
     image_input = torch.tensor(np.stack(images)).to(device)
     with torch.no_grad():
         image_features = model.encode_image(image_input)