[ECCV 2024]🎉 EntAugment: Entropy-Driven Adaptive Data Augmentation Framework for Image Classification

Introduction

This is the implementation of EntAugment and EntLoss, as used in the paper. In this paper, we propose a tuning-free and adaptive DA framework that dynamically assesses and adjusts the augmentation magnitudes for each sample during training. You can directly start off using our implementations.

Getting Started

• Codes support Python3

• Clone this directory and cd into it.

git clone https://github.com/Jackbrocp/EntAugment

cd EntAugment

Updates

• 2024/7/15: Initial release

Getting Started

Requirements

• Python 3
• PyTorch 1.6.0
• Torchvision 0.7.0
• Numpy

Run Data Augmentation

Prepare the datasets

Download the datasets (e.g., CIFAR datasets) and put the datasets under the folder data/

Parameters

--conf，path to the config file, e.g., confs/resnet18.yaml

Training Examples

Employ EntAugment as a data augmentation method to train ResNet18 model on CIFAR10 dataset.

python train_EntAugment.py --conf confs/resnet18.yaml --dataset CIFAR10

Employ EntAugment and EntLoss to train ResNet18 model on CIFAR100 dataset.

python train_EntLoss.py --conf confs/resnet18.yaml --dataset CIFAR100

Acknowledge

Part of our implementation is adopted from the TrivialAugment repositories.

Citation

If you find this repository useful in your research, please cite our paper:

@article{yang2024entaugment,
  title={EntAugment: Entropy-Driven Adaptive Data Augmentation Framework for Image Classification},
  author={Yang, Suorong and Shen, Furao and Zhao, Jian},
  journal={arXiv preprint arXiv:2409.06290},
  year={2024}
}