openai/CLIP

Trained on 400M image-text pairs using contrastive learning, CLIP jointly encodes images and text into a shared embedding space where cosine similarity enables zero-shot classification without task-specific fine-tuning. Built on Vision Transformers and text encoders in PyTorch, it integrates seamlessly with torchvision for preprocessing and supports multiple model scales (ViT-B/32, ViT-L/14, etc.) for deployment flexibility.

32,796 stars. Actively maintained with 1 commit in the last 30 days.