xinyu1205/recognize-anything

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

3708 stars · 327 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Recognize Anything is a collection of open-source image recognition foundation models, including RAM, RAM++, and Tag2Text, that perform image tagging and captioning with high accuracy. It provides pretrained models and inference code for recognizing thousands of common and open-set categories.

Use cases

• tag images with descriptive labels automatically
• recognize any common category in a photo without training
• generate captions for images with a tagging-guided model
• build a visual semantic analysis pipeline with Grounded-SAM
• classify images into open-set categories zero-shot

When to choose

• you need strong zero-shot image tagging or recognition
• you want open-source alternatives to proprietary tagging APIs like Google's
• you need tagging combined with captioning in one model
• you want to integrate recognition with segmentation models like SAM

When to avoid

• you need fine-grained classification on a small custom label set where a simple classifier suffices
• you have no GPU and need lightweight real-time inference
• you need object detection or localization alone rather than recognition

Facets

library · maturity active

image-processing computer-vision machine-learning deep-learning computer-vision image-processing artificial-intelligence machine-learning python cross-platform image-tagging zero-shot-recognition vision-language-model image-captioning ram tag2text foundation-model gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="xinyu1205/recognize-anything")