Relightable Detailed Human Reconstruction from Sparse Flashlight Images

Note: this repo is still under construction.

Usage

Create environment

git clone https://github.com/Jarvisss/Relightable_human_recon.git 
cd Relightable_human_recon
conda create -n relit_hmr python=3.9 && conda activate relit_hmr
conda install -c conda-forge igl
## install pytorch 2.0.1 with cu118
pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu118
## install kaolin
pip install kaolin==0.15.0 -f https://nvidia-kaolin.s3.us-east-2.amazonaws.com/torch-2.0.1_cu118.html
## install other dependencies
pip install -r requirements.txt

Training and testing

For pretraining on a human scan dataset:

bash pretrain.sh 

For joint optimization on custom data:

bash joint_optimize.sh

The full parameters are listed in options/options.py. A graphics card with at least 24 GB of memory is recommended. (E.g. RTX 3090)

Data

Please refer to the data folder for for information on the data structure.

Camera convention

We use the OpenGL camera convention for the rendering of synthetic data for the pretraining phase. Conversely, the OpenCV camera convention is adopted for the joint optimization phase.

Acknowledgements

We would like to thank the authors of the following works to open-source their wonderful projects.

• IDR
• NeuS
• PIFu
• IRON

Thanks for the open-sourcing of dataset Thuman 2.0