make sure the video have min(frame_count, 64) frames

inference.py

@@ -43,6 +43,7 @@ def center_crop(image):
 
     # read video
     cap, retaining, clips = cv2.VideoCapture(VIDEO_NAME), True, []
+    frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
     while retaining:
         retaining, frame = cap.read()
         if not retaining and frame is None:
@@ -55,7 +56,7 @@ def center_crop(image):
         tmp_ = center_crop(cv2.resize(frame, (resize_width, resize_height)))
         tmp = tmp_.astype(np.float32) / 255.0
         clips.append(tmp)
-        if len(clips) == clip_len:
+        if len(clips) == clip_len or len(clips) == frame_count:
             inputs = np.array(clips)
             inputs = np.expand_dims(inputs, axis=0)
             inputs = np.transpose(inputs, (0, 4, 1, 2, 3))

utils.py

@@ -163,7 +163,10 @@ def load_frames(self, file_dir):
     def crop(self, buffer, clip_len, crop_size):
         if self.split == 'train':
             # randomly select time index for temporal jitter
-            time_index = np.random.randint(buffer.shape[0] - clip_len)
+            if buffer.shape[0] > clip_len:
+                time_index = np.random.randint(buffer.shape[0] - clip_len)
+            else:
+                time_index = 0
             # randomly select start indices in order to crop the video
             height_index = np.random.randint(buffer.shape[1] - crop_size)
             width_index = np.random.randint(buffer.shape[2] - crop_size)
@@ -172,7 +175,10 @@ def crop(self, buffer, clip_len, crop_size):
             # jitter takes place via the selection of consecutive frames
         else:
             # for val and test, select the middle and center frames
-            time_index = math.floor((buffer.shape[0] - clip_len) / 2)
+            if buffer.shape[0] > clip_len:
+                time_index = math.floor((buffer.shape[0] - clip_len) / 2)
+            else:
+                time_index = 0
             height_index = math.floor((buffer.shape[1] - crop_size) / 2)
             width_index = math.floor((buffer.shape[2] - crop_size) / 2)
         buffer = buffer[time_index:time_index + clip_len, height_index:height_index + crop_size,