baaivision/EVA
EVA provides a series of vision foundation models including ViT variants, EVA-CLIP, and masked visual representation learning approaches from BAAI.
Velocity · 7d
+2.1
★ / day
Trend
→steady
star history
The EVA repository hosts multiple visual representation models from the Beijing Academy of Artificial Intelligence. It includes EVA-01 (masked visual representation learning at scale), EVA-02 (improved vision transformer architecture), and EVA-CLIP variants (contrastive language-image pretraining) scaling up to 18 billion parameters. The models are available through Hugging Face and timm libraries.