Skip to content

Fix first PTS bug #565

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Mar 17, 2025
Merged

Fix first PTS bug #565

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions src/torchcodec/decoders/_core/VideoDecoder.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -467,6 +467,7 @@ void VideoDecoder::addStream(
TORCH_CHECK_EQ(retVal, AVSUCCESS);

streamInfo.codecContext->thread_count = ffmpegThreadCount.value_or(0);
streamInfo.codecContext->pkt_timebase = streamInfo.stream->time_base;

// TODO_CODE_QUALITY same as above.
if (mediaType == AVMEDIA_TYPE_VIDEO && device.type() == torch::kCUDA) {
Expand Down
14 changes: 2 additions & 12 deletions test/decoders/test_decoders.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -990,13 +990,7 @@ def test_get_all_samples(self, asset, stop_seconds):
torch.testing.assert_close(samples.data, reference_frames)
assert samples.sample_rate == asset.sample_rate

# TODO there's a bug with NASA_AUDIO_MP3: https://github.com/pytorch/torchcodec/issues/553
expected_pts = (
0.072
if asset is NASA_AUDIO_MP3
else asset.get_frame_info(idx=0).pts_seconds
)
assert samples.pts_seconds == expected_pts
assert samples.pts_seconds == asset.get_frame_info(idx=0).pts_seconds

@pytest.mark.parametrize("asset", (NASA_AUDIO, NASA_AUDIO_MP3))
def test_at_frame_boundaries(self, asset):
Expand Down Expand Up @@ -1060,12 +1054,8 @@ def test_start_equals_stop(self, asset):
assert samples.data.shape == (0, 0)

def test_frame_start_is_not_zero(self):
# For NASA_AUDIO_MP3, the first frame is not at 0, it's at 0.072 [1].
# For NASA_AUDIO_MP3, the first frame is not at 0, it's at 0.138125.
# So if we request start = 0.05, we shouldn't be truncating anything.
#
# [1] well, really it's at 0.138125, not 0.072 (see
# https://github.com/pytorch/torchcodec/issues/553), but for the purpose
# of this test it doesn't matter.

asset = NASA_AUDIO_MP3
start_seconds = 0.05 # this is less than the first frame's pts
Expand Down
12 changes: 3 additions & 9 deletions test/decoders/test_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -826,6 +826,8 @@ def get_reference_frames(start_seconds, stop_seconds):

@pytest.mark.parametrize("asset", (NASA_AUDIO, NASA_AUDIO_MP3))
def test_pts(self, asset):
# Non-regression test for
# https://github.com/pytorch/torchcodec/issues/553
decoder = create_from_file(str(asset.path), seek_mode="approximate")
add_audio_stream(decoder)

Expand All @@ -840,15 +842,7 @@ def test_pts(self, asset):
frames, asset.get_frame_data_by_index(frame_index)
)

if asset is NASA_AUDIO_MP3 and frame_index == 0:
# TODO This is a bug. The 0.138125 is correct while 0.072 is
# incorrect, even though it comes from the decoded AVFrame's pts
# field.
# See https://github.com/pytorch/torchcodec/issues/553
assert pts_seconds == 0.072
assert start_seconds == 0.138125
else:
assert pts_seconds == start_seconds
assert pts_seconds == start_seconds


if __name__ == "__main__":
Expand Down
Loading