Recognize Anything

Recognize Anything Model (RAM) is the strong foundation model for zero-shot, high-accuracy image tagging across 6,400+ common categories.

RAM (Recognize Anything Model) delivers exceptional, open-set image tagging capabilities: it's a foundation model engineered for zero-shot generalization. This technology automatically recognizes over 6,400 common tags, significantly expanding the scope beyond datasets like OpenImages V6. We built RAM by leveraging large-scale image-text pairs and a four-step data engine, completely bypassing manual annotations. Performance metrics confirm its superiority: RAM significantly outperforms models like CLIP and BLIP, achieving competitive results even against the Google tagging API on numerous benchmarks.

https://recognize-anything.github.io/