FREQ-MIP-AA : FREQUENCY MIP REPRESENTATION FOR ANTI-ALIASING NEURAL RADIANCE FIELDS

Youngin Park, Seungtae Nam, Cheul-hee Hahm, Eunbyung Park*
*Corresponding author
International Conference on Image Processing (ICIP), 2024

Architecture overview

• The overall architecture of our model begins with training a shared grid in the frequency domain. This is followed by scale-specific low-pass filters designed to facilitate focus on crucial information. Subsequently, learnable frequency masks are applied to further refine frequency grids. To enhance visual clarity, the grid is shown as a basic square shape, even though it is fundamentally a vector matrix structure. The ⊙ represents element-wise multiplication.

Environment setup

• conda create -n freqMipAA python=3.8
• conda activate freqMipAA
• pip install -r requirements.txt

Training

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

• FreqMipAA on the multi-scale blender dataset.

bash ./scripts/multiscale_freqMipAA.sh

References

• Mip-Grid
• Tri-MipRF
• TensoRF