Evaluating the Impact of Data Augmentation on Image Classification

A reproducible, modular framework to compare the effect of various data augmentation techniques on image classification accuracy, reliability, and robustness using CNN, ResNet, and EfficientNet across CIFAR-10, MNIST, and ImageNet.

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/iAnisDev/data-augmentation-impact.git
cd data-augmentation-impact

2. Install dependencies

pip install -r requirements.txt

---

📂 Directory Structure

data-augmentation-impact/
├── data/
├── src/
│   ├── main.py
│   └── ...
├── experiments/
├── results/
├── README.md
├── requirements.txt
└── .gitignore

---

⚡️ Usage

All tasks (downloading data, preprocessing, training, evaluation) can be managed via the unified CLI interface.

Basic Command Format

python src/main.py [ACTION FLAGS] [OPTIONS]

Available Action Flags

Flag	Shortcut	Description
--load-data	--ld	Download datasets (CIFAR-10, MNIST, etc.)
--preprocess	--pp	Preprocess and clean data
--train	--tr	Train models (choose model/dataset/aug)
--evaluate	--ev	Evaluate models and output metrics/plots
--all	--a	Run the entire pipeline (load, preprocess, train, evaluate)
--augment	--aug	Specify augmentation strategy (traditional, advanced, gan, etc.)
--config	-c	Use experiment configuration file (YAML/JSON)
--help	-h	Show help message

---

🏁 Example Workflows

1. Download data only

python src/main.py --load-data

2. Preprocess data

python src/main.py --preprocess --dataset cifar10

3. Train a model (ResNet, with traditional augmentation, on CIFAR-10)

python src/main.py --train --model resnet --dataset cifar10 --augment traditional

4. Evaluate all models and output metrics/plots

python src/main.py --evaluate

5. Run the entire pipeline on MNIST with GAN augmentation

python src/main.py --all --dataset mnist --model efficientnet --augment gan

6. Use an experiment configuration file

python src/main.py --all --config experiments/configs/baseline_resnet_cifar10.yaml

---

🛠️ Arguments Reference

Argument	Description	Default
--dataset	Dataset to use (all,cifar10, mnist, imagenet)	all
--augment	Augmentation method (all, auto, traditional, miamix, mixup, lsb, vqvae, gan, fusion)	all
--model	Model to train (cnn, resnet, efficientnet)	cnn
--epochs	Number of training epochs	20
--batch-size	Batch size (For CPU and GPU)	64, 256
--config	Path to config file for advanced setups	optional

See python src/main.py --help for the full list of options and descriptions.

---

📊 Results

Outputs such as logs, metrics, and plots will be saved in the results/ directory.

---

📄 License

MIT License

---

📚 References

• List of all academic papers and links to datasets, see docs/references.md.