# NVlabs/describe-anything

[ICCV 2025] Implementation for Describe Anything: Detailed Localized Image and Video Captioning

Repository: https://github.com/NVlabs/describe-anything
Canonical: https://ross.abutalabs.com/products/describe-anything
Homepage: https://describe-anything.github.io/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: large-multimodal-models, vision-language-model, describe-anything, detailed-localized-captioning
Last push: 2025-06-26T03:57:51+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 28, release rhythm 35, longevity 36
- inputs: {"age_days": 516, "days_push": 433, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1514, forks 94 (observed 2026-08-28T04:04:56.319018+00:00)

## What it is
Describe Anything Model (DAM) is a vision-language model that generates detailed descriptions of user-specified regions in images and videos, indicated via points, boxes, scribbles, or masks. The repository provides the model implementation, Gradio demos, SAM integration, and the DLC-Bench evaluation benchmark.

## Use cases
- generate detailed captions for a specific region of an image
- describe objects in a video by annotating one frame
- caption image regions using masks from SAM
- evaluate localized captioning models with DLC-Bench
- build an app that describes clicked or boxed image areas
- run a local gradio demo for region-based image description

## When to choose
- you need fine-grained, localized image or video descriptions rather than whole-image captions
- you want a research-grade vision-language model with a permissive Apache-2.0 license
- you need a benchmark for evaluating detailed localized captioning

## When to avoid
- you only need simple whole-image captioning without region specification
- you lack a GPU or cannot run large multimodal models locally
- you need a production-ready hosted API rather than a research codebase

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, computer-vision, image-processing, video-processing, nlp
- domain: computer-vision, large-language-models, artificial-intelligence, image-processing
- platform: python, windows
- tags: vision-language-model, region-captioning, image-captioning, video-captioning, segmentation, gradio-demo, benchmark, iccv-2025, multimodal, video, linux, macos, gpu

## Member repositories
- NVlabs/describe-anything (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:56.319018+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:32:11.457163+00:00, confidence not recorded.
  - readme: https://github.com/NVlabs/describe-anything (fetched 2026-08-28T04:04:56.319018+00:00, sha 13e2a6651342)
  - homepage: https://describe-anything.github.io/ (fetched 2026-08-29T11:35:46.592583+00:00, sha 3f9b27456586)
- Data as of 2026-08-30T08:39:29.467469+00:00.