XeniaLLL/Prior-LT

Towards Calibrated Model for Long-Tailed Visual Recognition from Prior Perspective

Zhengzhuo Xu, Zenghao Chai, Chun Yuan

This is the PyTorch implementation of our paper in NeurIPS 2021.

Environment settings

PyTorch >= 1.4
Scikit-learn
Matplotlib

Training

To train the model, please select a config file path or customize by yourself. For example:

python train.py config/cifar10_100.py

The result will be saved in ./result.

Inferance model

Dataset	log	Model
CIFAR-10-LT-50	link	link
CIFAR-10-LT-100	link	link
CIFAR-10-LT-200	link	link
CIFAR-100-LT-50	link	link
CIFAR-100-LT-100	link	link
CIFAR-100-LT-200	link	link

Note

Some settings may be a little different from the paper reported. Because we make further optimization considering reviewers' suggestions.
Some bugs need to be fixed when imb factor = 0.1
Need to solve the test-agnostic situation.

Reference

Welcome to cite:

@inproceedings{
    xu2021towards,
    title={Towards Calibrated Model for Long-Tailed Visual Recognition from Prior Perspective},
    author={Zhengzhuo Xu and Zenghao Chai and Chun Yuan},
    booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
    year={2021},
    url={https://openreview.net/forum?id=vqzAfN-BoA_}
}

Acknowledgement

We adopt the code from the following repos. We thank them for providing their awesome code.

XeniaLLL / Prior-LT