# shubham-goel/4D-Humans

4DHumans: Reconstructing and Tracking Humans with Transformers

Repository: https://github.com/shubham-goel/4D-Humans
Canonical: https://ross.abutalabs.com/products/4d-humans
Homepage: https://shubham-goel.github.io/4dhumans/
Language: Python
License: MIT
License Family: permissive
Topics: 3d-reconstruction
Last push: 2026-02-07T16:46:00+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 66, release rhythm 35, longevity 85
- inputs: {"age_days": 1190, "days_push": 207, "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 1671, forks 176 (observed 2026-08-28T04:05:20.112654+00:00)

## What it is
4DHumans is a Python research codebase implementing HMR 2.0, a transformer-based model for 3D human mesh recovery from single images, plus a PHALP-based system for reconstructing and tracking humans in 3D from monocular video. It is the official code for the ICCV 2023 paper 'Humans in 4D: Reconstructing and Tracking Humans with Transformers'.

## Use cases
- reconstruct 3d human meshes from photos
- track multiple people across video frames in 3d
- estimate human pose and shape from a single image
- extract smpl meshes from video for animation or analysis
- run human action recognition features from pose
- process in-the-wild videos with occlusions and unusual poses

## When to choose
- you need state-of-the-art single-image 3D human mesh recovery
- you want to track people's identities through occlusion in monocular video
- you need SMPL-format meshes as output for downstream research
- you want a research baseline for pose-based action recognition

## When to avoid
- you need real-time performance on edge devices
- you require a production-ready supported product rather than research code
- you cannot register for and obtain the SMPL model files
- you need multi-camera or depth-sensor input rather than monocular RGB

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, computer-vision, image-processing
- domain: computer-vision, machine-learning, deep-learning, artificial-intelligence
- platform: python
- tags: 3d-reconstruction, human-mesh-recovery, human-pose-estimation, transformers, hmr2, human-tracking, smpl, monocular-video, research-code, iccv-2023, linux, macos, gpu

## Member repositories
- shubham-goel/4D-Humans (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:20.112654+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 2026-08-30T03:42:42.543503+00:00, confidence not recorded.
  - readme: https://github.com/shubham-goel/4D-Humans (fetched 2026-08-28T04:05:20.112654+00:00, sha c0cadc7fa2b2)
  - homepage: https://shubham-goel.github.io/4dhumans/ (fetched 2026-08-29T11:15:50.456213+00:00, sha c082e4f01958)
- Data as of 2026-08-30T08:39:29.467469+00:00.