README.md

Use the Detection Transformer as a Data Augmenter

PyTorch implementation of DeMix | paper link

Method Overview

Setup

Install Package Dependencies

pip install -r requirements.txt

Datasets

download the fine-grained datasets:

CUB-200-2011

Stanford-Cars devkit, train, test, test_annos_withlabels

FGVC-Aircraft

DETR object detection

1, import the function:from datasets.dataset_process import compute_detr_res

2, run the function: compute_detr_res(dataset_name='cub', datadir='cub data dir') # ['cub', 'car', 'aircraft']

Training

python demix.py
    --dataset='cub' # ['cub', 'car', 'aircraft']
    --netname='resnet18' # ['resnet18', 'resnet34', 'resnet50', 'resnet101', 'inception_v3', 'densenet121']
    --mixmethod='detrmix' # ['detrmix', 'saliencymix', 'mixup', 'cutmix']
    --pretrained=1 # if training from scratch, set pretrained=0

Citation

If you find this code useful, please kindly cite

@article{wang2023use,
  title={Use the Detection Transformer as a Data Augmenter},
  author={Wang, Luping and Liu, Bin},
  journal={arXiv preprint arXiv:2304.04554},
  year={2023} 
}

Acknowledgment

This code is based on the SnapMix.

Contact

If you have any questions or suggestions, please feel free to contact wangluping/[email protected].