Spaces:
Running
Running
File size: 1,132 Bytes
7c930e7 3172ebe d08621a 3172ebe |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 |
---
title: README
emoji: 😻
colorFrom: blue
colorTo: purple
sdk: static
pinned: false
---
## About
**SynthFairCLIP** is a research initiative focused on fair vision–language models.
We study how to reduce bias in CLIP-style models by combining:
- **Real data** from large-scale datasets such as DataComp/CommonPool.
- **Synthetic data** generated with state-of-the-art diffusion models.
- **Curation and balancing** of demographic attributes across professions, activities and contexts.
---
### What we release
- **CLIP models** trained on hybrid real–synthetic data.
- **Large-scale WebDataset shards** of synthetic / hybrid image–text data.
- **Eval tools and benchmarks** for analysing bias and fairness in CLIP-like models.
[](https://github.com/lluisgomez/SynthFairCLIP/tree/main/evals)
If you use our resources, please consider citing the SynthFairCLIP project.
---
### Acknowledgement
We acknowledge EuroHPC JU for awarding the project ID EHPC-AI-2024A02-040 access to MareNostrum 5 hosted at BSC-CNS. |