Our Siggraph'18 paper on MC denoising with CNNs is out. It's a follow up to our paper from Sig'17 that addresses temporal stability, lightweight adaptation/retraining, detail preservation via training with asymmetric losses, etc. drz.disneyresearch.com/~jnovak/public…

Happy to announce that we are open-sourcing JERI--a JavaScript viewer for EXR images. It can be embedded into a website and provides interactive viewing experience. Took almost a year to get it out... see live demos here jeri.io

Our new paper on denoising deep images is online. Make no mistake, there is no deep learning in there. In this one we rely on the good old NL-Means filter in combination with a deep-z, cross-bilateral filter operating on auxiliary features. zurich.disneyresearch.com/~jnovak/public…

Our SIGGRAPH paper is finally out. It's all about delivering image quality of the past using rendering techniques of the future. Or as one reviewer put it: "this paper shows how to render remarkably poor images in a fascinating way". jannovak.info/publications/C…

Our work on Neural Control Variates has been accepted to SIGGRAPH Asia 2020. It's about online-trained neural nets that approximate outgoing radiance and help with correcting the approximation error. Handles RGB better than importance sampling alone. jannovak.info/publications/N…

In case you wonder what a rendering paper without any rendered images looks like (any images actually), here is an example.

Sharing our latest SIGGRAPH paper on neural scene representations -- a new controllable and scalable representation inspired by traditional scene graphs. It allows intuitive artistic control and can be used to produce temporally stable animations. developer.nvidia.com/blog/nvidia-re…

We are excited to share our ACM SIGGRAPH 2021 paper "An unbiased ray-marching transmittance estimator" (with @ejdeon, Jacopo Pantaleoni, Jan Novák) introducing two new MC transmittance estimators that can be 10-100x more efficient than prior methods such as ratio tracking. (1/7)

We finally managed to obtain an approval for releasing some of the training data used in our SIGGRAPH'17 denoising paper (KPCN). Might be handy if you want to train your own deep-learning denoiser. Disclaimer: no production data, only renders of CC scenes. disneyresearch.com/publication/de…

We are releasing another dataset. This time it is a bunch volumes stored as Field3D and VDB files that Simon Kallweit procedurally generated for the Deep Scattering project. disneyresearch.com/publication/de…

Slides for our EG state-of-the-art report about MC methods for light transport in volumes are online drz.disneyresearch.com/~jnovak/public…