Seungtae Nam, Daniel Rho, Jong Hwan Ko, Eunbyung Park*
*Corresponding author
in Neural Information Processing Systems (NeurIPS), 2023

• Mip-Grid generates multi-scale grids by applying convolution to a single-scale grid with learnable kernels.
• Two multi-scale grids closest to given scales are interpolated.

• please follow the official document for installation.
• if you already installed both of them, please skip this part.

• run the command below at "/your/path/to/MipGrid".
• don't forget to include the dot . at the end.

docker build -t mipgrid_environment .

• run the command below at "/your/path/to/MipGrid".
• please also change the dataset path ("/your/path/to/dataset").

docker run -it \
    -v ./:/workspace \
    -v /hdd/stnamjef/dataset:/workspace/dataset \
    --ipc host \
    --gpus all \
    --name mipgrid \
    mipgrid_environment:latest

• run the commands below at "/your/path/to/mipTensoRF" inside the docker container.

• original TensoRF on the multi-scale Blender dataset.

bash ./scripts/blender/tensor_vm.sh

• single-scale TensoRF on the multi-scale Blender dataset.

bash ./scripts/blender/singlescale_tensor_vm.sh

• multi-scale TensoRF on the multi-scale Blender dataset.

bash ./scripts/blender/multiscale_tensor_vm.sh

• mipTensoRF (disc.) on the multi-scale Blender dataset.

bash ./scripts/blender/mip_tensor_vm_discrete.sh

• mipTensoRF (cont.) on the multi-scale Blender dataset.

bash ./scripts/blender/mip_tensor_vm_continuous.sh

• mipTensoRF (2D) on the multi-scale Blender dataset.

bash ./scripts/blender/mip_tensor_vm_2d.sh