splash.webm

This repository contains the code for foot reconstruction using dense correspondences, as shown in our paper:

FOCUS - Multi-View Foot Reconstruction from Synthetically Trained Dense Correspondences
3DV 2025
Oliver Boyne and Roberto Cipolla
[arXiv] [project page]

Set-up

1. git clone --recurse-submodules https://github.com/OllieBoyne/FOCUS.git
2. Install FOCUS and dependencies: pip install --editable .¹
3. Download any required data (see Downloads section).
4. If using uncalibrated images, install COLMAP.

Usage

FOCUS can be run using the FOCUS/run_focus.py script:

• Use the --method flag to choose between sfm (FOCUS-SfM) and o (FOCUS-O).
• For a directory of images, use the --source_folder flag.
• For a video, use the --video_path flag.

See the run_focus.py script for more options.

Downloads

Item	Description	Download to
TOC model	A pre-trained correspondence predictor model.	data/toc_model
Foot 3D	Multi-view foot dataset for evaluating method.	data/Foot3D
3D Fits	3D reconstructions for evaluation, as used in the paper.	data/3d_fits

Citation

If you use our work, please cite:

@inproceedings{boyne2025focus,
            title={FOCUS - Multi-View Foot Reconstruction from Synthetically Trained Dense Correspondences},
            author={Boyne, Oliver and Cipolla, Roberto},
            booktitle={2025 International Conference on 3D Vision (3DV)},
            year={2025}
}

Footnotes

1. Use environment variable SKIP_PYTORCH3D=1 if you need to install PyTorch3D separately (e.g. in Windows) ↩