Anny-Fit

Official code for Anny-Fit: All-age Human Mesh Recovery accepted at CVPR 2026 Findings
Laura Bravo-Sánchez, Matthieu Armando, Romain Brégier, Grégory Rogez, Serena Yeung-Levy, Fabien Baradel

Our method recovers multi-person 3D human meshes from all ages directly in camera space. By integrating expert semantic, depth, keypoint, and segmentation cues, it improves all-age HMR and enables zero-shot adaptation of adult-only models.

News

• 206/05/05 — Code released

---

Installation

git clone --recurse-submodules https://github.com/naver/anny-fit
bash scripts/install.sh
bash scripts/download_checkpoints.sh
source setup.sh

See INSTALL.md for details and troubleshooting tips.

---

Demo

Download the pre-processed demo data (images + preprocessing) for running Anny-Fit on sample images:

wget -O demo_data.tar.gz https://download.europe.naverlabs.com/ComputerVision/AnnyFit/demo_data.tar.gz
tar xzf demo_data.tar.gz

Or run the preprocessing from scratch on a folder of images using a Multi-HMR 3D mesh initialization (requires all checkpoints):

python -m preprocess.build_test_dataset \
    --data_root demo \
    --dataset_name multi_person \
    --preprocess_data \
    --detector detectron2

Visualize the preprocessing:

# Optional: visualize preprocessing for a random sample from a folder
python -m visualize.visualize_preprocessing --data_root demo -n 10

# Fitting with Anny-Fit:
cd annyfit
python optimize.py --config configs/demo/multihmr.yaml

---

License

Code is provided under the terms of this LICENSE and accompanying NOTICE.

---

Citation

If you find our paper or code useful you can cite our work with:

@inproceedings{anny-fit2026,
    title={Anny-Fit: All-age Human Mesh Recovery},
    author={Bravo-S{\'a}nchez, Laura and
            Armando, Matthieu and
            Br{\'e}gier, Romain and 
            Rogez, Gr{\'e}gory and
            Yeung-Levy, Serena and
            Baradel, Fabien
            },
    booktitle={CVPR Findings},
    year={2026}
}

---

Our other works

Check out our other works that made this paper possible:

• Anny a differentiable all-age human body model.
• Anny-One a synthetic dataset of 780K+ multi-person and multi-view images with Anny ground-truth meshes.
• Multi-HMR a regression-based 3D human mesh estimation model.