the frame variable is not getting updated in the for loop in the frames_from_video_file function, resulting in an ndarray containing the same frame n_frames times.
I took the liberty to rewrite the for loop
for i in range(1, n_frames):
src.set(cv2.CAP_PROP_POS_FRAMES, start + frame_step * i)
ret, frame = src.read()
if ret:
result.append(format_frames(frame, output_size))
else:
result.extend([np.zeros_like(result[0]) for j in range(n_frames - i)])
break