Original implementation for NeurIPS 2020 paper Stochastic Normalization.
[News] 2022/04/27 Please refer to the Transfer Learning Library for a modular implementation.
- python3
- torch == 1.1.0 (with suitable CUDA and CuDNN version)
- torchvision == 0.3.0
- numpy
- argparse
- tqdm
| Dataset | Download Link |
|---|---|
| CUB-200-2011 | http://www.vision.caltech.edu/visipedia/CUB-200-2011.html |
| Stanford Cars | http://ai.stanford.edu/~jkrause/cars/car_dataset.html |
| FGVC Aircraft | http://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/ |
| NIH Chest X-ray | https://nihcc.app.box.com/v/ChestXray-NIHCC |
python --gpu [gpu_num] --data_path /path/to/dataset --class_num [class_num] --p 0.5 train.py
We re-trained our StochNorm with this code on full 15% train data of CUB-200-2011. Results are shown in the table below.
| Sampling Rate | Top-1 Acc(%) |
|---|---|
| 15% | 50.41 |
| 30% | 62.09 |
| 50% | 72.05 |
| 100% | 79.65 |
If you use this code for your research, please consider citing:
@article{kou2020stochastic,
title={Stochastic Normalization},
author={Kou, Zhi and You, Kaichao and Long, Mingsheng and Wang, Jianmin},
journal={Advances in Neural Information Processing Systems},
volume={33},
year={2020}
}
If you have any problem about our code, feel free to contact kz19@mails.tsinghua.edu.cn.