Frame number mismatch.
Thanks for your excellent work!
When I am processing the dataset, I found a mismatch in the frame number between the num_frames {number of video frames N (scalar)} and the actual extracted frame number from videos. This mismatch encountered for almost half of the samples I have processed.
Specifically,
- For the
num_frames {number of video frames N (scalar)}, it is obtained from the downloaded.pklfile. - For
the actual extracted frame number from videos: I firstly download the YouTube videos using the command in Video download, and then extract frames using the command in Video frame extraction. Every step is conducted in a standard approach as listed in the readme.
Do you have any suggestions in solving this problem?
Thanks for your help in advance!
Best regards,
Yihang
Hi Yihang,
Thanks for your interest and for sharing this issue! Can you be a bit more specific about the type of mismatch? Are the number of frames off by one compared to the expected? Are there far fewer frames being downloaded than expected?
I think looking into the timestamps for the clip can tell us the expected number of frames.
- If the actual number of frames matches this, and N is close, perhaps we can clip beginning or ending frames to make things match (maybe there is some rounding issue difference)
- If the actual number of frames is much less, it would seem there is some issue with either frame extraction or not enough of the full video is being downloaded
- If N is much less or much more, perhaps there has been another operation performed between extraction and final processing
Please let me know what you run into! Some initial tests for me did not see this issue. You said about half of your examples have this issue. Could it also be a certain part of the dataset that seems to have this?
Best,
Chris