FARE4 Convnext Base W Laion2b S13b B82k Augreg

🚀 Open Clip

A robust perceptual metric based on the CLIP model, offering high - performance in zero - shot image classification.

[Paper] [GitHub]

This is a robust perceptual metric, which is based on the CLIP model laion/CLIP - convnext_base_w - laion2B - s13B - b82K - augreg. It has been adversarially fine - tuned with FARE (Schlarmann et al. (2024)) on ImageNet with infinity - norm and radius 4/255.

Performance on the perceptual similarity task NIGHTS

Clean     L-inf, eps=4/255     L2, eps=3
90.6      74.3                 66.1

🚀 Quick Start

The following shows how to use this model.

💻 Usage Examples

Basic Usage

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

📄 License

This project is licensed under the MIT license.

📚 Documentation

Citation

If you find this model useful, please consider citing our papers:

@inproceedings{croce2024adversarially,
  title={Adversarially Robust CLIP Models Can Induce Better (Robust) Perceptual Metrics},
  author={Croce, Francesco and Schlarmann, Christian and Singh, Naman Deep and Hein, Matthias},
  year={2025},
  booktitle={{SaTML}}
}

@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}}
}