---
library_name: pytorch
tags:
- opensdi
- maskclip
- diffusion-detection
- image-forensics
- forgery-localization
- pytorch
datasets:
- nebula/OpenSDI_train
- nebula/OpenSDI_test
---

# MaskCLIP Weights for OpenSDI

This repository hosts model checkpoints for **OpenSDI: Spotting Diffusion-Generated Images in the Open World**.

## Links

- Model weights: https://huggingface.co/nebula/MaskCLIP-weights/tree/main
- Code: https://github.com/iamwangyabin/OpenSDI
- Project page: https://iamwangyabin.github.io/OpenSDI/
- Paper: https://arxiv.org/abs/2503.19653
- Training dataset: https://huggingface.co/datasets/nebula/OpenSDI_train
- Testing dataset: https://huggingface.co/datasets/nebula/OpenSDI_test

## Checkpoints

The `Files and versions` tab contains `.pth` checkpoints for MaskCLIP and related OpenSDI baselines. For MaskCLIP evaluation, download one of the `MaskCLIP_sd15_*.pth` checkpoints and use it with the OpenSDI codebase.

Example:

```bash
hf download nebula/MaskCLIP-weights MaskCLIP_sd15_20241103_17_45_16.pth --local-dir weights
```

Then set `--checkpoint_path` in `test.sh` to the downloaded checkpoint path, for example:

```bash
--checkpoint_path "weights/MaskCLIP_sd15_20241103_17_45_16.pth"
```

## Citation

If you find OpenSDI useful for your research and applications, please cite:

```bibtex
@InProceedings{wang2025opensdi,
    author={Wang, Yabin and Huang, Zhiwu and Hong, Xiaopeng},
    title={OpenSDI: Spotting Diffusion-Generated Images in the Open World},
    booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    year={2025}
}
```