This repo contains the code for the real-time application of DIVeR implemented in Python+CUDA.

• python 3.8
• pytorch 1.9.0
• moderngl
• moderngl-window

We have two versions of pre-trained models trained on nerf_synthetic dataset.

• Our 256 model (256x256x256 voxel grid).
• Our 128 model (128x128x128 voxel grid).

The 128 model runs much faster with smaller storage cost, but the 256 model has better rendering quality.

To launch the real-time application, run:

python run.py --weight_path=PATH_TO_WEIGHT_FILE \
			  --voxel_num=VOXEL_GRID_SIZE \
			  --device=GPU_DEVICE

• Project page
• Paper
• Training code

@misc{wu2021diver,
      title={DIVeR: Real-time and Accurate Neural Radiance Fields with Deterministic Integration for Volume Rendering}, 
      author={Liwen Wu and Jae Yong Lee and Anand Bhattad and Yuxiong Wang and David Forsyth},
      year={2021},
      eprint={2111.10427},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}