JIA-Lab-research/LISA

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

2674 stars · 206 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

LISA (Large Language Instructed Segmentation Assistant) is a multimodal large language model that performs reasoning-based image segmentation, producing segmentation masks from complex natural language queries. It combines an LLM with a segmentation model to answer questions like 'segment the object that...' while explaining its reasoning.

Use cases

• segment objects in an image described by complex reasoning queries
• generate segmentation masks from natural language instructions
• build a multimodal assistant that explains what it segments
• fine-tune a vision-language model for segmentation tasks
• run a local demo of LLM-driven image segmentation
• research reasoning segmentation with the LISA++ dataset

When to avoid

• you only need classic promptable segmentation like SAM with point or box prompts
• you lack GPU resources for large multimodal model inference
• you need a lightweight production segmentation pipeline without LLM overhead

Facets

library · maturity active

machine-learning deep-learning computer-vision image-processing llm-inference nlp computer-vision large-language-models machine-learning artificial-intelligence image-processing python cross-platform segmentation multimodal vision-language-model reasoning-segmentation gradio-demo research linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="JIA-Lab-research/LISA")