# Arthur151/ROMP

Monocular, One-stage, Regression of Multiple 3D People and their 3D positions & trajectories in camera & global coordinates. ROMP[ICCV21], BEV[CVPR22], TRACE[CVPR2023]

Repository: https://github.com/Arthur151/ROMP
Canonical: https://ross.abutalabs.com/products/romp
Homepage: https://www.yusun.work/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: 3d-mesh-recovery, bottom-up, multi-person-3d-mesh-recovery, smpl, multi-person, pytorch, pose-estimation
Last push: 2024-11-14T06:04:07+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 2203, "days_push": 657, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1538, forks 246 (observed 2026-08-28T04:05:00.275848+00:00)

## What it is
ROMP is a PyTorch-based library and pip-installable API (simple-romp) for real-time monocular multi-person 3D human mesh recovery, implementing the ROMP (ICCV21), BEV (CVPR22), and TRACE (CVPR23) models. It regresses SMPL body meshes, depth relationships, subject tracking, and global 3D trajectories from single-camera video, with export to fbx/glb/bvh formats.

## Use cases
- recover 3d human meshes from a single image or video
- estimate multi-person 3d poses and positions from monocular camera footage
- track specific people across video frames and get their global 3d trajectories
- estimate depth relationships between multiple people in a scene
- export 3d human motion to fbx glb or bvh for animation pipelines
- run real-time 3d pose estimation on gpu or cpu with onnx
- drive blender or game-engine characters from video of people

## When to choose
- you need multi-person 3d body mesh recovery from monocular video in real time
- you want SMPL-format outputs with global coordinates and camera tracking
- you need cross-platform (Linux/Windows/Mac) inference via pip with fbx/glb/bvh export
- you want to reproduce or extend published research models (ROMP, BEV, TRACE)

## When to avoid
- you need high-accuracy single-person pose with heavy occlusion handling beyond these models' scope
- you require face or hand mesh recovery (SMPL-X style full-body)
- you need a maintained production product - development activity has slowed since 2023-2024
- you work outside Python/PyTorch ecosystems and cannot use the provided API or Docker

## Facets
- artifact type: library
- maturity: stable
- function: computer-vision, machine-learning, deep-learning, image-processing
- domain: computer-vision, machine-learning, deep-learning, artificial-intelligence
- platform: windows, python
- tags: 3d-human-pose-estimation, smpl, multi-person, mesh-recovery, pytorch, human-tracking, 3d-trajectory, onnx, linux, macos, gpu

## Member repositories
- Arthur151/ROMP (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:00.275848+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 2026-08-30T04:30:55.610537+00:00, confidence not recorded.
  - readme: https://github.com/Arthur151/ROMP (fetched 2026-08-28T04:05:00.275848+00:00, sha 29f1996d1dcb)
  - homepage: https://www.yusun.work/ (fetched 2026-08-29T11:32:34.749987+00:00, sha 9130fa6ebde3)
- Data as of 2026-08-30T08:39:29.467469+00:00.