Adversarial Monte Carlo Denoising with Conditioned Auxliary Feature Modulation - Test Suite Overview
The following scenes were rendered with the Tungsten renderer and denoised using several different apporaches.
All techniques use auxiliary buffers specified by their papers and code. KPCN denotes [Bako et al.2017]; RAE denotes [Chaitanya et al.2017]; NFOR denotes [Bitterli et al.2016].
Results for other techniques are rendered using their public released code and weights. Results by RAE [Chaitanya et al.2017] are denoised using their public executable since their code is not released. The results include input rendered spanning 4, 16, 32, 64, 128 sample per pixel. Evaluation metrics include SSIM, PSNR, RMSE.
Acknowledgements
We would like to thank NFOR for their dataset and interactive viewer.