UA-Fit (HRNet-W40)
Learn 2D, Solve 3D: UA-Fit, an Uncertainty-Weighted Analytical Solver for Multi-view Hand Mesh Recovery. ECCV 2026 Workshops (HANDS). Marsil Zakour, Constantin Patsch, Martin Piccolrovazzi, Yuankai Wu, Eckehard Steinbach, Technical University of Munich.
Project page 路 Paper (PDF) 路 Code 路 Solver (parafit)
The network predicts per-view 2D evidence (a dense joint vote field with a per-vote 2x2 precision), triangulates it with its uncertainty, and an analytical Levenberg-Marquardt solver fits a parametric hand to it. The same weights fit MANO (the paper), Meta's MHR hand or the UmeTrack hand.
Files
| file | content |
|---|---|
uafit_w40.safetensors |
network weights (HRNet-W40 backbone, vote-field / uncertainty / shape heads), 189 MB |
The file contains only the network. It does not contain MANO model data: download MANO from https://mano.is.tue.mpg.de/ (its own license) as described in the code repository.
Usage
git clone https://github.com/MarsilZakour/ua-fit-code && cd ua-fit-code
pip install -r requirements.txt
python scripts/evaluate.py --checkpoint Marsil123/ua-fit # downloads uafit_w40.safetensors
python scripts/evaluate.py --checkpoint Marsil123/ua-fit --hand-model mhr
from uafit.network import load_uafit_network
network = load_uafit_network("Marsil123/ua-fit", device="cuda")
Results (Table 1, full official test splits, mm)
| dataset (views) | MPVPE | MPJPE | PA-MPVPE |
|---|---|---|---|
| DexYCB (8) | 7.86 | 7.55 | 4.92 |
| HO3D (5) | 8.01 | 7.41 | 5.10 |
| ARCTIC (8) | 7.54 | 6.62 | 5.66 |
| InterHand2.6M (8) | 8.77 | 7.79 | 6.23 |
| OakInk (4) | 9.58 | 8.86 | 5.67 |
Reproduced with these weights and the released code (scripts/evaluate.py).
Training data and license
Trained on multi-view DexYCB, HO3D, ARCTIC, InterHand2.6M, OakInk and FreiHAND. Several of these are released for non-commercial research only; the weights are therefore released under CC BY-NC 4.0 for research use.
Citation
@inproceedings{zakour2026uafit,
title = {Learn 2D, Solve 3D: UA-Fit, an Uncertainty-Weighted Analytical Solver for Multi-view Hand Mesh Recovery},
author = {Zakour, Marsil and Patsch, Constantin and Piccolrovazzi, Martin and Wu, Yuankai and Steinbach, Eckehard},
booktitle = {European Conference on Computer Vision (ECCV) Workshops -- HANDS},
year = {2026}
}