Setup anaconda environment using environment.yml
file.
conda env create --name DiffuseMix --file=environment.yml
conda remove -n DiffuseMix --all # In case environment installation faileds
train
└─── class 1
└───── n04355338_22023.jpg
└─── ...
val
└─── class 1
└───── n03786901_5410.jpg
└─── ...
To introduce the structural complexity, you can download fractal image dataset from here Fractal Dataset
`python3 main.py --train_dir PATH --fractal_dir PATH --prompts sunset,Autumn
Method | Stanford Cars |
---|---|
Vanilla(CVPR'16) | 85.52 |
RA(NIPS'20) | 87.79 |
AdaAug(ICLR'22) | 88.49 |
PuzzleMix(ICML'20) | 89.68 |
Co-Mixup(ICLR'21) | 89.53 |
Guided-AP(AAAI'23) | 90.27 |
DiffuseMix | 91.26 |
Accuracy of the network on the 8041 test images: 91.23%
If you find our work useful in your research please consider citing our paper:
@article{diffuseMix2024,
title={DIFFUSEMIX: Label-Preserving Data Augmentation with Diffusion Models},
author={Khawar Islam, Muhammad Zaigham Zaheer, Arif Mahmood, Karthik Nandakumar},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2024}
}