New release from Valeo.ai: Franca, a high-performing, fully open-source vision foundation model (data, code, weights). It matches and sometimes outperform DINOv2/SigLIPv2/CLIP on ViT-G, setting a new open-source standard. GitHub.