Fix decoding issue with PYAV due to new support for multiple training…

[dfan]

See issue #181. When using pyav as a backend, we get the exception pyav with exception: unsupported operand type(s) for -: 'list' and 'int'. This is because num_frames is now a list. torchvision_decode() is correct but not pyav_decode().

[facebook-github-bot]

Hi @dfan!

Thank you for your pull request and welcome to our community.

Action Required

In order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you.

Process

In order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA.

Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with CLA signed. The tagging process may take up to 1 hour after signing. Please give it that time before contacting us about it.

If you have received this in error or have any questions, please contact us at cla@fb.com. Thanks!

[facebook-github-bot]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

[facebook-github-bot]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks!

[facebook-github-bot]

@lyttonhao has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[XinyuSun]

Hi! There is an error in the code:
Line 442 and 443 should be:
442: video_start_pts = int(start_idx * timebase)
443: video_end_pts = int(end_idx * timebase)

[facebook-github-bot]

@lyttonhao has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[dfan]

Very sorry ... yes you are right. With this additional fix I am now able to achieve correct results with the new code. Thanks and sorry again!

[lyttonhao]

Very sorry ... yes you are right. With this additional fix I am now able to achieve correct results with the new code. Thanks and sorry again!

Do you want to update the code? Thanks!

[junchen14]

See issue #181. When using pyav as a backend, we get the exception pyav with exception: unsupported operand type(s) for -: 'list' and 'int'. This is because num_frames is now a list. torchvision_decode() is correct but not pyav_decode().

this can solve the problems that I faced

[StawEndl]

Very sorry ... yes you are right. With this additional fix I am now able to achieve correct results with the new code. Thanks and sorry again!

hi,
i use the pyav_decode functiopn to read the video, but this function return the decode_all_video which is False. When decode_all_video is False, the start_end_delta_time is not created.
#557
how to fix it?

[XinyuSun]

pyav is much slower than other decode backends. I'm considering adding support to the decord backend.

[facebook-github-bot]

@lyttonhao has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[facebook-github-bot]

@dfan has updated the pull request. You must reimport the pull request before landing.

[facebook-github-bot]

@lyttonhao has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[lyttonhao]

Looks great! We should also handle the case when duration is None

[lyttonhao]

For line 414-417, we should also decode the whole videos and return the frames. Example code could be:

        decode_all_video = True
        video_start_pts, video_end_pts = 0, math.inf
        start_end_delta_time = None

        frames = None
        if container.streams.video:
            video_frames, max_pts = pyav_decode_stream(
                container,
                video_start_pts,
                video_end_pts,
                container.streams.video[0],
                {"video": 0},
            )

            frames = [frame.to_rgb().to_ndarray() for frame in video_frames]
            frames = torch.as_tensor(np.stack(frames))
        frames_out = [frames]

[facebook-github-bot]

@dfan has updated the pull request. You must reimport the pull request before landing.

[dfan]

Good catch! I saw there were additional fixes to the decoder.py in 6069ea9 - do you want me to include those changes or will you merge those in?

[lyttonhao]

Good catch! I saw there were additional fixes to the decoder.py in 6069ea9 - do you want me to include those changes or will you merge those in?

Yeah, that would be great if you could also include those changes!

[facebook-github-bot]

@dfan has updated the pull request. You must reimport the pull request before landing.

[dfan]

Apologies for the delay

[cir7]

@dfan hi, i am curious why remove parameter use_offset when call get_multiple_start_end_idx in line 456

[M-Masuhara]

Hi. I started using SlowFast last week and I get the same error as you.

As for DECODING_BACKEND, I have tried pyav and torchvision, but both show the following error.

pyav:
Failed to decode by pyav with exception: unsupported operand type(s) for -: 'list' and 'int'
Failed to decode video idx 63 from C:/ ~ video path ; trial 99

torchvision:
Failed to decode by torchvision with exception: 'OpNamespace' object has no attribute 'probe_video_from_memory'
Failed to decode video idx 63 from C:/ ~ video path ; trial 99

As for slowfast>datasets>decoder.py, it has already been corrected as described here.

The libraries version are as follows:

PyTorch = 1.12.1
torchvision = 0.13.1
ffmpeg = 1.4
pyav = 9.2.0

What is not good?
I would like to solve this problem somehow with pyav.
Please help.

[zsz00]

I meet leaks issue too.