InstantHMR recovers 3D human meshes in real time on phones
Original title: InstantHMR : Real-Time Human Mesh Recovery on Your Phone
A Galaxy S23 NPU produces a 6,890-vertex mesh in 5.5 ms, with 40.0 mm PA-MPJPE on 3DPW.
InstantHMR is an open-source family of single-image human mesh regressors aimed at mobile inference. Models use the Momentum Human Rig and are trained only on public SAM 3D Body annotations. The authors report 5.5 ms per person on a Samsung Galaxy S23 NPU and 0.98 ms on an NVIDIA RTX 4070, and say this is faster than the methods they evaluated. They re-evaluated 11 methods under a unified protocol and measured latency on eight targets.
Unconfirmed: Speed and accuracy claims come from the authors' own evaluation protocol.
- Mobile accuracy is 40.0 mm PA-MPJPE on 3DPW, 48.8 mm on EMDB, and 41.7 mm on RICH.
- On mobile it is 12x faster than the one compatible method that is slightly more accurate.
- Latency was measured from desktop TensorRT to two phone NPUs.
- Code and models are at github.com/Naturalpad-studio/InstantHMR.
- Organizations
- Meta
- AI models
- InstantHMR
From the original article
1 min read at huggingface.co
Real-time 3D human mesh recovery from a single image on mobile devices is a challenging problem with many applications, including robotics, biomechanics, AR/VR, and exergames. Current state-of-the-art models are either too slow to run efficiently on mobile devices or sacrifice accuracy for speed.
By Mohamed Adjel, Kahina Chalabi, Vincent Bonnet · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.