Hello. I’m using the VIA pipeline for a video understanding task. When I try to process a video longer than 4 hours, it gets stuck and cannot preprocess any further. After that, I attempted to create multiple instances of the decoder component in this pipeline to preprocess several videos simultaneously. However, I still encounter the same issue. I suspect this error is related to queue management. Can you help me solve this problem?
What chunk size are you using to process the video? Can you try increasing the chunk size to a point where it does not crash? Try 5 and 10 minute chunks.
I have already tried with chunks 10m or 5m, and it can run. Also, i customized the decoder code and the videoframegetter class. I get only the videoframegetter class and i found that with each instance, if i try get_frame method for more than about1000 chunkInfo, it can not run anymore. I try my code in A30 gpu.
VIA does not currently support A30. As per the documentation, the GPUs supported are A600, L40, L40s, A100, and H100.
Thank you. But when i tried this code in A100, it also faced the same problem.
I thought that this stuck is in “self._loop.run()” line in VideoFileFrameGetter class.
Maybe it ran out of some resources after you changed the code. Could you run the nvidia-smi dmon command to check the resource usage when you run your command?
Thank you. I used a timeout to quit this loop and prevent it from getting stuck.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.