rtcamp8-denoiser

レイトレ合宿8のデノイズ部門用に作成したデノイザーです。

Requirements

• C++ 17
• CUDA 11.7
• CMake 3.20

Features

• Bilateral filter
• NL-Means filter
• A-trous filter
• Nonlinearly weighted first-order regression filter

Build

mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j8

Run

./main <filepath-to-color.hdr> <filepath-to-albedo.hdr> <filepath-to-normal.hdr>

References

• Dammertz, H., Sewtz, D., Hanika, J. and Lensch, H.P., 2010, June. Edge-avoiding a-trous wavelet transform for fast global illumination filtering. In Proceedings of the Conference on High Performance Graphics (pp. 67-75).
• Bitterli, B., Rousselle, F., Moon, B., Iglesias‐Guitián, J.A., Adler, D., Mitchell, K., Jarosz, W. and Novák, J., 2016, July. Nonlinearly weighted first‐order regression for denoising Monte Carlo renderings. In Computer Graphics Forum (Vol. 35, No. 4, pp. 107-117).
• Schied, C., Kaplanyan, A., Wyman, C., Patney, A., Chaitanya, C.R.A., Burgess, J., Liu, S., Dachsbacher, C., Lefohn, A. and Salvi, M., 2017. Spatiotemporal variance-guided filtering: real-time reconstruction for path-traced global illumination. In Proceedings of High Performance Graphics (pp. 1-12).