HMR 2.0
Fully transformerized human mesh recovery from a single image — the backbone of the 4D-Humans tracking system.
About
HMR 2.0 is the modern descendant of the classic HMR human mesh recovery network, rebuilt around a Vision Transformer backbone with a transformer decoder head. The upgrade matters: it handles unusual poses and viewpoints that the 2018-era HMR fumbled, and it anchors 4DHumans, the full system that runs HMR 2.0 per frame and stitches detections into multi-person 3D tracks across video — surviving occlusions that break frame-by-frame methods.
The ICCV 2023 paper reports state-of-the-art results across 3DPW, Human3.6M, LSP-Extended and PoseTrack, and it also proved useful downstream — the authors showed it improving pose-based action recognition. The repo (MIT) auto-downloads checkpoints on first run, and the community has ported it into everything from Colab demos to sports-analysis pipelines. Like most SMPL-based work, you register once on the SMPL site to grab the body model.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Paper (arXiv 2305.20091): state-of-the-art results on 3DPW, Human3.6M, LSP-Extended, PoseTrack (author-reported)
- Powers 4D-Humans monocular multi-person tracking; ICCV 2023
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (TIER verified, metrics-v1). Evidence: ICCV 2023 paper (arXiv 2305.20091) with multi-dataset evals; MIT-licensed official repo (shubham-goel/4D-Humans) with ~1.7k stars; checkpoints auto-download; widely used as a backbone in human-tracking work; project page live. Strengths: peer-reviewed, permissive license, strong adoption. Docked for: (1) headline results are author-reported on academic sets; (2) requires a separately registered SMPL model download. License: MIT.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · ICCV 2023 paper (arXiv 2305.20091), Humans in 4D; authors at UC Berkeley
- · Official repo shubham-goel/4D-Humans, MIT, ~1.7k stars; checkpoints auto-download on first run
- · Paper claims SOTA on 3DPW, Human3.6M, LSP-Extended, PoseTrack; project page live