search

jayleicn / moment_detr

[NeurIPS 2021] Moment-DETR code and QVHighlights dataset
https://arxiv.org/abs/2107.09609
MIT License
267 stars 44 forks source link

Experiments on Charades-STA dataset #22

Closed minjoong507 closed 1 year ago

minjoong507 commented 1 year ago

Hi there, thanks for sharing your great work!

Following issues #11 and #8, I'm trying to train the model on Charades-STA.

However, I got the error message 'raise ValueError("Sample larger than population or is negative")'.

I think that the model cannot sample the negative clips because some gt moments are covering the entire video.

https://github.com/jayleicn/moment_detr/blob/1e67364fb55c9a74b08a3e02824adfea622adb94/moment_detr/start_end_dataset.py#L104-L117

Thanks.

onlyonewater commented 1 year ago

@minjoong507 , when I train the model on the charades-sta dataset, I meet the same error, have you solved it?

minjoong507 commented 1 year ago

Hi @onlyonewater

Actually, it's been a while since I solved this problem. So I don't remember exactly what caused this problem.

In Charades-STA, some noisy annotations have longer moment length than video duration.

I'm not sure, but these cases may arise the problem because you can't sample negative clips from them.

onlyonewater commented 1 year ago

hi, @minjoong507, I have already solved this problem, if the time of moment annotation is as long as the duration of the whole video, the error may appear, because, in this case, the neg_pool has fewer elements than the number that wants to sample. So, I filtered these annotations, and the error did not appear.