search

ArrowLuo / CLIP4Clip

An official implementation for "CLIP4Clip: An Empirical Study of CLIP for End to End Video Clip Retrieval"
https://arxiv.org/abs/2104.08860
MIT License
851 stars 121 forks source link

MSRVTT dataset #1

Closed BlueCat7 closed 3 years ago

BlueCat7 commented 3 years ago

Dear author, thanks your work. When I download MSRVTT dataset, there are some urls broken. How do you solve it, thanks.

ArrowLuo commented 3 years ago

@BlueCat7 The videos are fine when we download them. It is a normal issue encountering urls broken. Keep the same test split is enough to make the comparison fair. I think it is ok if the missing ratio is low.

BlueCat7 commented 3 years ago

Thank you for your quick response to my issue. But I miss about 2 thousand videos, maybe it's a problem to training.

ArrowLuo commented 3 years ago

It is indeed a problem. Do you check the original URL? Is the linked video removed? If this is the case, you can check the overlap between the training (MSRVTT_train.9k.csv or MSRVTT_train.7k.csv), test splits (MSRVTT_JSFUSION_test.csv), and your downloaded videos. How much they lost respectively?

BlueCat7 commented 3 years ago

Ok, I will check it firstly, thanks.

ArrowLuo commented 3 years ago

If you want to compare the results with baselines including ours, you can rerun the baselines on your dataset. To CLIP4Clip, replace MSRVTT_train.9k.csv or MSRVTT_train.7k.csv and MSRVTT_JSFUSION_test.csv with yours CSV will be okay. It is a compromise though will cost a long time.

BlueCat7 commented 3 years ago

Ok, thank you for your suggestion.