zer0int/CLIP-fine-tune-cross-attention-read-noread — GitHub trending stats & insights

zer0int/CLIP-fine-tune-cross-attention-read-noread

A cross-attention bridge for CLIP, to read OR ignore text in images on-demand. For typographic attack robustness or deliberately leveraging CLIP's 'OCR-like' text-reading capabilities.

Visit GitHub

Python

1 contributors

MIT License

Social mentions

Recent discussions about this repository across the web

To Read AND Not To Read. At the flip of a switch - with a cross-attention bridge! CLIP typographic attacks be gone. Praised be the reading CLIP. Hugging Face model, full training code, paper is ~60%…

@zer0int1 · x.com

Repository activities

repository's daily and monthly activities across stars, forks, merged PRs, issues, and closed issues

Data is not available yet

Recent activity data for stars, forks, merged PRs, issues, and closed issues will appear here once available