Let Gaussians Deform with Scale-specific Mipmap for Anti-aliasing Rendering.
Clone the repository and create an anaconda environment using
git clone https://github.com/renaissanceee/Mipmap-GS.git
cd Mipmap-GS
conda create -y -n mipmap-gs python=3.8
conda activate mipmap-gs
pip install torch==1.12.1+cu113 torchvision==0.13.1+cu113 -f https://download.pytorch.org/whl/torch_stable.html
conda install cudatoolkit-dev=11.3 -c conda-forge
pip install -r requirements.txt
pip install submodules/diff-gaussian-rasterization
pip install submodules/simple-knn
Please download and unzip nerf_synthetic.zip from the NeRF's official Google Drive.
Please download the data from the Mip-NeRF 360 and request the authors for the treehill and flowers scenes.
Download SwinIR checkpoints into here.
# single-scale training and multi-scale testing (stmt) on nerf-synthetic
python scripts/run_syn_stmt_up.py --->zoom-in
python scripts/run_syn_stmt_down.py --->zoom-out
# mip-nerf 360
python scripts/run_360_stmt_up.py
python scripts/run_360_stmt_down.py
# zoom-in mipmap
python pseudo_GT/SwinIR/run_swinir_syn.py --source "./nerf_synthetic/" --gt_root "./benchmark_nerf_synthetic_stmt_up"
python pseudo_GT/SwinIR/run_swinir_360.py
# zoom-out mipmap
python pseudo_GT/bilinear/downsample_syn.py --source "./nerf_synthetic/" --gt_root "./benchmark_nerf_synthetic_stmt_down"
python pseudo_GT/bilinear/downsample_360.py
# run Mipmap-GS on nerf-synthetic
python scripts/ours/run_syn_stmt_up_swin.py --->zoom-in
python scripts/ours/run_syn_stmt_down.py --->zoom-out
# mip-nerf 360
python scripts/ours/run_360_stmt_up_swin.py
python scripts/ours/run_360_stmt_down.py
# metrics
python scripts/ours/metrics_syn_stmt_up_swin.py
python scripts/ours/metrics_syn_stmt_down.py
python scripts/ours/metrics_360_stmt_up_swin.py
python scripts/ours/metrics_360_stmt_down.py
This project is built upon 3DGS and Mip-Splatting. The SR model is from SwinIR. We thank all the authors for their great work and repos.