How to drop/skip some frames and decode without decoding the drop frames?

[PythonImageDeveloper]

I have a python script that can read bytes of a long video. I want to skip some frames in each second. for example the fps of the video is 24 and I want to drop those frames that are not multiplication by 3 if the frame numbers are like this list [1,2,3,4, ...,24] , and I expect this list [3,6,9,12,15,...,24]

the problem is that, if I skip the frames before decoding in pyav, the returned frames are corrupted (when I saved them I see some noises in the saved frames) and also it gave me an error. here is my python code:

` stream_options = [{'codec': 'h264'}]
container = av.open(video_path, stream_options=stream_options)

video_stream = [s for s in container.streams if s.type == "video"][0]
packet_list = []
for id_p, packet in enumerate(container.demux(video_stream)):
    packet_list.append(packet)

skip=3
for id_pack, packet in enumerate(packet_list):
    if id_pack % skip==0:
        frame = packet.decode()
        if len(frame):
            frame_np = frame[0].to_ndarray(format='rgb24')
            frame_list.append(frame_np)`

it gave me an error as below:

co located POCs unavailable reference picture missing during reorder reference picture missing during reorder Missing reference picture, default is 65636

Is there any way to drop frames before decoding in pyav? decoding all frames takes a long time.

[ghylander]

I'm interested in this too (left a message in gitter too)

My goal is similar:

• open a stream, don't decode the frames
• when certain condition is met, grab and decode a single video frame (the current, real-time frame, i.e. the last frame that entered)

[jlaine]

For most formats I think this is going to be impossible: at best packets which contain keyframe are going to be decodable on their own. For other packets, decoding the frame requires decoding the previous packets first.

To be clear: this is not a limitation of FFmpeg or PyAV, this is just how encoding works.

[ghylander]

That was insightful, thanks. You are 100% right and this was a total oversight on my side.

Most codecs use the information in consecutives frames to transmit the data (if some part of the next frame does not change from the current one, don't send that data and re-use the already sent one).

Real shame this approach is so difficult.