# hkchengrex/Tracking-Anything-with-DEVA

[ICCV 2023] Tracking Anything with Decoupled Video Segmentation

Repository: https://github.com/hkchengrex/Tracking-Anything-with-DEVA
Canonical: https://ross.abutalabs.com/products/tracking-anything-with-deva
Homepage: https://hkchengrex.com/Tracking-Anything-with-DEVA/
Language: Python
License: NOASSERTION
License Family: other
Topics: deep-learning, object-tracking, open-vocabulary-segmentation, video-editing, video-object-segmentation, video-segmentation, open-vocabulary-video-segmentation, open-world-video-segmentation, iccv2023
Last push: 2025-04-26T19:46:00+00:00

## Health v2 (maintenance only)
Score: 27/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 18, release rhythm 8, longevity 79
- inputs: {"age_days": 1112, "days_push": 494, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1508, forks 142 (observed 2026-08-28T04:04:55.531633+00:00)

## What it is
DEVA is a decoupled video segmentation framework that combines task-specific image-level segmentation models with a universal bi-directional temporal propagation model to produce coherent video segmentation. It supports open-vocabulary, text-prompted tracking of arbitrary objects in videos and can integrate custom image segmentation models without finetuning.

## Use cases
- segment and track objects in videos with text prompts
- open-vocabulary video segmentation without task-specific video training data
- propagate image segmentation masks across video frames
- apply Segment Anything or GroundingDINO to long videos
- video object segmentation for research benchmarks like DAVIS
- integrate a custom image segmentation model into video tracking

## When to choose
- you need open-vocabulary or open-world video segmentation with text prompts
- you want to reuse an image-level segmentation model on videos without finetuning
- you need long-term coherent tracking of many objects in research settings

## When to avoid
- you need a lightweight real-time production video pipeline
- you lack a GPU or cannot run large deep-learning models
- you need a turnkey end-user application rather than a research codebase

## Facets
- artifact type: library
- maturity: stable
- function: computer-vision, image-processing, machine-learning, deep-learning
- domain: computer-vision, image-processing, artificial-intelligence, deep-learning
- platform: python, cross-platform
- tags: video-segmentation, object-tracking, open-vocabulary, segment-anything, video-object-segmentation, research, iccv-2023, pytorch, video, linux, gpu

## Member repositories
- hkchengrex/Tracking-Anything-with-DEVA (main) score 27

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:55.531633+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 2026-08-30T04:32:33.117948+00:00, confidence not recorded.
  - readme: https://github.com/hkchengrex/Tracking-Anything-with-DEVA (fetched 2026-08-28T04:04:55.531633+00:00, sha 79304a24b6bf)
  - homepage: https://hkchengrex.com/Tracking-Anything-with-DEVA/ (fetched 2026-08-29T11:36:54.667224+00:00, sha 5a93ad206e0f)
- Data as of 2026-08-30T08:39:29.467469+00:00.