🎨 DISTILLATION-STYLE-CNN

Use style-transfer to boost continual learning!

🌟 Overview

This project implements the paper: STCR: Enhancing Generalization for Continual Learning through Style Transfer and Content Replay

🛠️ Getting Started

Prerequisites

• 🐍 Python
• 📦 Pip

Installation

1. Clone the repo:

   git clone https://github.com/jingcjie/distillation-style-cnn
   

2. Navigate to the project directory:

   cd distillation-style-cnn
   

3. Install dependencies:

   pip install -r requirements.txt
   

Usage

Run the main script with:

python main_imagenet.py [arguments]

Key arguments include:

• --batch-size: Batch size (default: 256)
• --epochs1: Number of epochs for first phase (default: 70)
• --epochs2: Number of epochs for second phase (default: 40)
• --new-classes: Number of new classes (default: 10)
• --start-classes: Number of initial classes (default: 50)
• --dataset: Choose 'imagenet100' or 'imagenet'
• --exp-name: Experiment name
• --no_STCR: Disable STCR
• --no_distill: Disable distillation loss
• --style_type: Style transfer method (0: adain, 1: traditional)

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

I will update this later if needed.

🙏 Citation

Arxiv version (waiting TNNLS bibtext):

@misc{shi2024unveilingtapestryinterplaygeneralization,
      title={Unveiling the Tapestry: the Interplay of Generalization and Forgetting in Continual Learning}, 
      author={Zenglin Shi and Jing Jie and Ying Sun and Joo Hwee Lim and Mengmi Zhang},
      year={2024},
      eprint={2211.11174},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2211.11174}, 
}