Zhengzhuo Xu, Shuo Yang, Xingjun Wang, Chun Yuan

This repository is the official PyTorch implementation of the paper in ICASSP 2023.  

We propose Prediction Distribution Calibration (PDC) to quantitatively evalate how proposals overcome model head preferance in LTR.

from eval import evaluate_all_metric

model, dataloader = None, None
# build your owen cls mode and image dataloader
device = 'cpu'
cls_num = None # list
# get train label distribution
res = evaluate_all_metric(
    dataloader,
    model,
    device,
    cls_num
)
pdc = res['pdc']

We will release the training code with ViTs soon.

If you find our idea or code inspiring, please cite our paper:

@inproceedings{PDC,
  author       = {Xu, Zhengzhuo and
                  Yang, Shuo and
                  Wang, Xingjun and
                  Yuan, Chun},
  title        = {Rethink Long-Tailed Recognition with Vision Transformers},
  booktitle    = {{IEEE} International Conference on Acoustics, Speech and Signal Processing, {ICASSP} 2023},
  year         = {2023}
}