mrSDMUniPS

Meshroom plugin for SDM-UniPS (CVPR2023) -- universal photometric stereo for surface normal and BRDF estimation.

Quick Start

Prerequisite: a working Meshroom installation (2025+).

1. Clone the plugin

cd /path/to/your/plugins
git clone https://github.com/meshroomHub/mrSDMUniPS.git
cd mrSDMUniPS

2. Set up the virtual environment

Meshroom looks for a folder named venv at the plugin root.

python3 -m venv venv
source venv/bin/activate

pip install --upgrade pip
pip install torch torchvision
pip install -r requirements.txt

deactivate

This installs SDM-UniPS and all its dependencies automatically via pip.

3. Download pretrained weights

bash download_weights.sh

This downloads the pretrained checkpoints (~480 MB) from Dropbox into checkpoint/:

checkpoint/
├── normal/
│   └── nml.pytmodel
└── brdf/
    └── brdf.pytmodel

The plugin auto-detects this directory. No config.json needed.

4. Register the plugin in Meshroom

export MESHROOM_PLUGINS_PATH=/path/to/your/plugins/mrSDMUniPS:$MESHROOM_PLUGINS_PATH

Launch Meshroom: the SDMUniPS node appears under Photometric Stereo.

Usage

Input

• SfMData JSON with multi-lighting views grouped by poseId (multiple images per pose, each under different lighting)
• Mask folder (optional) with mask PNGs named by poseId (e.g. 12345.png)

Output

• Normal maps -- surface normals per pose (<poseId>_normals.png)
• Albedo maps -- base color per pose (<poseId>_albedo.png)
• Roughness maps -- surface roughness per pose (<poseId>_roughness.png)
• Metallic maps -- metallic parameter per pose (<poseId>_metallic.png)
• Output SfMData files -- normalMaps.sfm, albedoMaps.sfm, roughnessMaps.sfm, metallicMaps.sfm

Advanced: Developer Setup

If you prefer to work from a local SDM-UniPS clone instead of pip install:

1. Clone the repo: git clone -b meshroom https://github.com/meshroomHubWarehouse/SDM-UniPS-CVPR2023.git
2. Edit meshroom/config.json:

   [
       {"key": "SDM_UNIPS_PATH", "type": "path", "value": "/path/to/SDM-UniPS-CVPR2023"},
       {"key": "SDM_UNIPS_CHECKPOINT_PATH", "type": "path", "value": "/path/to/SDM-UniPS-CVPR2023/checkpoint"}
   ]

The node tries pip imports first, then falls back to the config path.

References

Ikehata, S. "Scalable, Detailed and Mask-free Universal Photometric Stereo." CVPR 2023.