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

Reproduction on LSMDC DataSet #105

Open JingXiaolun opened 1 year ago

JingXiaolun commented 1 year ago

Sorry to disturb you. When I reproduce the results on LSMDC dataset, I get worse results than those in paper. In the meanP experiment, the meanR is always around 200, rather than around 60. The log is shown as follows: 1686898632103 The problem has confused me for a long time and I hope to get some advice for addressing this issue

JingXiaolun commented 1 year ago

@ArrowLuo

shallowdream66 commented 1 year ago

I have observed your training logs and reproduced it. I found that the performance was poor only on the validation set, but returned to normal on the test set. I hope this will be helpful to you

JingXiaolun commented 10 months ago

@shallowdream66 ，the test set contains 1000 video clips, and the validation set contains 7408 video clips, do you mean using the 7408 video clips in validation process ?

shallowdream66 commented 10 months ago

Yes，after each epoch it will be verified on the validation set, the checkpoint with the best results will be taken for testing, and the test results will return to normal.

JingXiaolun commented 10 months ago

@shallowdream66, thanks a lot, I will try it.

JingXiaolun commented 10 months ago

@shallowdream66 , can you share the training LSMDC log with me ? I still can't reproduce the normal results in the test set