ACE

ACE: Ally Complementary Experts for Solving Long-Tailed Recognition in One-Shot

This repository is the official PyTorch implementation of ICCV-21 paper ACE: Ally Complementary Experts for Solving Long-Tailed Recognition in One-Shot.

Prerequirements

To install the environment.

conda env create -f environment.yml
conda activate ACE
git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

• Data format

The annotation of a dataset is a dict consisting of two field: annotations and num_classes. The field annotations is a list of dict with image_id, fpath, im_height, im_width and category_id.

Here is an example.

{
    'annotations': [
                    {
                        'image_id': 1,
                        'fpath': '/data/iNat18/images/train_val2018/Plantae/7477/3b60c9486db1d2ee875f11a669fbde4a.jpg',
                        'im_height': 600,
                        'im_width': 800,
                        'category_id': 7477
                    },
                    ...
                   ]
    'num_classes': 8142
}

Usage

Training

#bash data_parallel_train.sh configuration_file_path GPU_indexes
bash data_parallel_train.sh configs/cifar100_im100.yaml 0,1 

Testing

#python valid.py configuration_file_path
python valid.py configs/cifar100_im100.yaml

Trained models

Acknowledgement

This project is developed based on Bag of tricks @AAAI-21, thanks for their works!

Citation

@inproceedings{cai2021ace,
  title={ACE: Ally Complementary Experts for Solving Long-Tailed Recognition in One-Shot},
  author={Cai, Jiarui and Wang, Yizhou and Hwang, Jenq-Neng},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={112--121},
  year={2021}
}