README.md

---
license: mit
library_name: open_clip
pipeline_tag: zero-shot-image-classification
---
[[Paper]](https://openreview.net/forum?id=e3scLKNiNg&noteId=e3scLKNiNg) [[GitHub]](https://github.com/fra31/perceptual-metrics)

Robust perceptual metric, based on CLIP model `laion/CLIP-convnext_base_w-laion2B-s13B-b82K-augreg`

Adversarially fine-tuned with FARE ([Schlarmann et al. (2024)](https://arxiv.org/abs/2402.12336)) on ImageNet with infinity-norm and radius 4/255.

Performance on the perceptual similarity task [NIGHTS](https://dreamsim-nights.github.io):
```
Clean     L-inf, eps=4/255     L2, eps=3
90.6      74.3                 66.1
```

## Usage
```python
model, _, image_processor = open_clip.create_model_and_transforms('hf-hub:chs20/FARE4-convnext_base_w-laion2B-s13B-b82K-augreg')
```

## Citation
If you find this model useful, please consider citing our papers:
```bibtex
@inproceedings{croce2024adversarially,
  title={Adversarially Robust CLIP Models Induce Better (Robust) Perceptual Metrics},
  author={Croce, Francesco and Schlarmann, Christian and Singh, Naman Deep and Hein, Matthias},
  year={2024},
  booktitle={{ICML Workshop on Foundation Models in the Wild}}
}
```

```bibtex
@inproceedings{schlarmann2024robustclip,
    title={Robust CLIP: Unsupervised Adversarial Fine-Tuning of Vision Embeddings for Robust Large Vision-Language Models}, 
    author={Schlarmann, Christian and Singh, Naman Deep and Croce, Francesco and Hein, Matthias},
    year={2024},
    booktitle={{ICML}}
}
```