xyzhang77 / volsdf-reproduce

VolSDF-reproduce

This repo reproduces VolSDF, and its official implementation is here.

TODO

• Implement background rendering.
• Optimize training process to avoid OOM and improve the utilization of graphics memory.

Dataset

It's only support DTU dataset now. Please refer to the procedure in official github.

Usage

train

python train.py --config ./config/DTU.txt \{--pretrained_model CKPT\}

--pretrained_model is used to resume training.

test

python train.py --config ./config/DTU.txt \{--pretrained_model CKPT\} --scan_id SCAN_ID --render_only

Citation

@inproceedings{yariv2021volume,
  title={Volume rendering of neural implicit surfaces},
  author={Yariv, Lior and Gu, Jiatao and Kasten, Yoni and Lipman, Yaron},
  booktitle={Thirty-Fifth Conference on Neural Information Processing Systems},
  year={2021}
}