Why the result of test video is very long?

dingfengshi / TriDet

[CVPR2023] Code for the paper, TriDet: Temporal Action Detection with Relative Boundary Modeling

MIT License

160 stars 13 forks source link

Why the result of test video is very long? #30

Open zhangfusheng-163 opened 8 months ago

zhangfusheng-163 commented 8 months ago

Hello,I have testing one video, the length of the video is 33 seconds，but the result is very long .I output the result：dict_keys(['video-id', 't-start', 't-end', 'label', 'score']),but the value length of 'video_id' or 'label' is 276000.Is it because each frame of the image has multiple results?

dingfengshi commented 8 months ago

Did you test on your dataset? The predicted segments will be clamped in the range of video length (see here). It could be that your config file is incorrectly set.

zhangfusheng-163 commented 7 months ago

@dingfengshi ,thanks.I had been resolved the problem。

But when I used Epic to load my own dataset, the results were very poor. Can you give me some suggestions for improvement?

dingfengshi commented 7 months ago

Due to the different types of videos in various datasets, it is challenging to determine the transfer effects. It is recommended to retrain a model on your own dataset.