PatchSAE: Sparse Autoencoders Reveal Selective Remapping of Visual Concepts During Adaptation

🚀 Quick Navigation

• Getting Started
• Interactive Demo
• Training & Analysis
• Status Updates
• License & Credits

🛠 Getting Started

Set up your environment with these simple steps:

# Create and activate environment
conda create --name patchsae python=3.12
conda activate patchsae

# Install dependencies
pip install -r requirements.txt

# Always set PYTHONPATH before running any scripts
cd patchsae
PYTHONPATH=./ python src/demo/app.py

🎮 Interactive Demo

Online Demo on Hugging Face 🤗

Explore our pre-computed images and SAE latents without any installation!

💡 The demo may experience slowdowns due to network constraints. For optimal performance, consider disabling your VPN if you encounter any delays.

Local Demo: Try Your Own Images

Want to experiment with your own images? Follow these steps:

1. Setup Local Demo

First, download the necessary files:

• out.zip
• data.zip

You can download the files using gdown as follows:

# Activate environment first (see Getting Started)

# Download necessary files (35MB + 513MB)
gdown --id 1NJzF8PriKz_mopBY4l8_44R0FVi2uw2g  # out.zip
gdown --id 1reuDjXsiMkntf1JJPLC5a3CcWuJ6Ji3Z  # data.zip

# Extract files
unzip data.zip
unzip out.zip

💡 Need gdown? Install it with: conda install conda-forge::gdown

Your folder structure should look like:

patchsae/
├── configs/
├── data/      # From data.zip
├── out/       # From out.zip
├── src/
│   └── demo/
│       └── app.py
├── tasks/
├── requirements.txt
└── ... (other files)

2. Launch the Demo

PYTHONPATH=./ python src/demo/app.py

⚠️ Note:

• First run will download datasets from HuggingFace automatically (About 30GB in total)
• Demo runs on CPU by default
• Access the interface at http://127.0.0.1:7860 (or the URL shown in terminal)

📊 PatchSAE Training and Analysis

• Training Instructions: See tasks/README.md
• Analysis Notebooks:
  • demo.ipynb
  • analysis.ipynb

📝 Status Updates

• Jan 13, 2025: Training & Analysis code work properly. Minor error in data loading by class when using ImageNet.
• Jan 09, 2025: Analysis code works. Updated training with evaluation during training, fixed optimizer bug.
• Jan 07, 2025: Added analysis code. Reproducibility tests completed (trained on ImageNet, tested on Oxford-Flowers).
• Jan 06, 2025: Training code updated. Reproducibility testing in progress.
• Jan 02, 2025: Training code incomplete in this version. Updates coming soon.

📜 License & Credits

Reference Implementations

• SAE for ViT
• SAELens
• Differentiable and Fast Geometric Median in NumPy and PyTorch
• Self-regulating Prompts: Foundational Model Adaptation without Forgetting [ICCV 2023]
  • Used in: configs/ and msrc/models/
• MaPLe: Multi-modal Prompt Learning CVPR 2023
  • Used in: configs/models/maple/...yaml and data/clip/maple/imagenet/model.pth.tar-2

License Notice

Our code is distributed under an MIT license, please see the LICENSE file for details. The NOTICE file lists license for all third-party code included in this repository. Please include the contents of the LICENSE and NOTICE files in all re-distributions of this code.

---

Citation

If you find our code or models useful in your work, please cite our paper:

@inproceedings{
  lim2025sparse,
  title={Sparse autoencoders reveal selective remapping of visual concepts during adaptation},
  author={Hyesu Lim and Jinho Choi and Jaegul Choo and Steffen Schneider},
  booktitle={The Thirteenth International Conference on Learning Representations},
  year={2025},
  url={https://openreview.net/forum?id=imT03YXlG2}
}