mx-mark / VideoTransformer-pytorch

PyTorch implementation of a collections of scalable Video Transformer Benchmarks.
281 stars 35 forks source link

What is the final score of maskfeat? #21

Closed daidaiershidi closed 2 years ago

mx-mark commented 2 years ago

@daidaiershidi We have no exact score on Kinetics but have pretrained on ImageNet1k, reaching a similar score as the original paper (ViT-B: 84.0, ViT-L: 85.9).

daidaiershidi commented 2 years ago

Could you please release the code on imagenet? I tried to reproduce 300 epochs of pre-training and 100 epochs of finetune and only got a top1 score of 81.6, much lower than the score in the paper (83.6). I don't know what detail is wrong, it's confusing... ...

mx-mark commented 2 years ago

@daidaiershidi It is belongs to a part of a project. We will release the code in the next few weeks.

RachelTeamo commented 2 years ago

I'm looking forward to the code on Imagenet. When will it be released? Thanks a lot.

daidaiershidi commented 2 years ago

@RechelTeamo The official project has been released. https://github.com/facebookresearch/SlowFast/blob/main/projects/maskfeat/README.md

RachelTeamo commented 2 years ago

Thanks again!