search

Megvii-BaseDetection / DisAlign

Implementation of "Distribution Alignment: A Unified Framework for Long-tail Visual Recognition"(CVPR 2021)
Apache License 2.0
117 stars 10 forks source link

Can someone kindly share their codes of Classification task on ImageNet_LT? #17

Closed smallcube closed 2 years ago

smallcube commented 2 years ago

I tried to train the proposed method on ImageNet_LT, but I can only get an average testing rate about 49%, which is far from the rate described in the paper (52.9). Some of the details regarding my implementations are given as follows: (1) The feature extractor is ResNexT-50 and the head classifier is a linear classifier. The testing accuracy in Stage-One is 43.9%, which is OK.

(2) The testing accuracy of adopting cRT method in Stage-Two is 49.6%, which is identical to one reported in other papers. (3) When fine-tuning the model in Stage-2, both the feature-extractor and head classifier are frozen, and a DisAliLinear model (which is implemented in CVPODs) is retrained. The testing accuracy can only reach 48.8%, which is far away from the one reported in your paper.

qianlanwyd commented 2 years ago

If you get a bad stage-1 model, you cannot get a good stage-2 model.

In fact, the stage-1 model DisAlign used is much better than other repos(e.g., Decoupling Representation and Classifier for Long-Tailed Recognition)

So, the comparison is not fair enough.

smallcube commented 2 years ago

If you get a bad stage-1 model, you cannot get a good stage-2 model.

In fact, the stage-1 model DisAlign used is much better than other repos(e.g., Decoupling Representation and Classifier for Long-Tailed Recognition)

So, the comparison is not fair enough.

Thanks for the feedback. By now, I think the performance of Stage-1 model is OK. So I was wondering if you can kindly share more details regarding the implementation of DisAlign, especially the part about calculating the KL divergence. I would appreciate that.

tonysy commented 2 years ago

@smallcube Hi, the implementation of the KL divergence is here: https://github.com/Megvii-BaseDetection/DisAlign/blob/5943553d8fdff41719a24d31fd5a58e676032cde/classification/imagenetlt/resnext50/resx50.scratch.imagenet_lt.224size.90e.disalign.10e/net.py#L50

smallcube commented 2 years ago

@smallcube Hi, the implementation of the KL divergence is here:

https://github.com/Megvii-BaseDetection/DisAlign/blob/5943553d8fdff41719a24d31fd5a58e676032cde/classification/imagenetlt/resnext50/resx50.scratch.imagenet_lt.224size.90e.disalign.10e/net.py#L50

Thanks for the feedback.