Denoising with Kernel Prediction and Asymmetric Loss Functions
Subscribe:  iCal  |  Google Calendar
Kiev UA   8, Dec — 10, Dec
Marina Del Rey US   10, Dec — 13, Dec
Las Vegas US   8, Jan — 12, Jan
Zürich CH   31, Jan — 4, Feb
Leamington Spa GB   31, Jan — 3, Feb
Latest comments
by Duacan
5 hours ago

hello Alexander, I really loved your these draw works. I loved cathedrals too.I started 3ds Max new. And I really really want to meet you, if you wanna to do. By the way, my name is Duacan, from Turkey. also Im working for learning and speaking German. Cause Deutschland is the my first country for living. Whatever, take care yourself, Tschüss. insta: 06optimusprime06

by Asadullah Sanusi
1 days ago

nice blog but here is the thing, what is wrong with overlaping uv's and mirroring them, what are the cons of overlapping them and why is this method better in the case of uv? thanks

Thank you @Fcardoso The volumetric light is available in the latest 2018.3 beta. In the visual environment setting, there is a new option to select Volumetric light fog. The screen I shared is from 2018.2 during that time I was using a script to enable it :)

Denoising with Kernel Prediction and Asymmetric Loss Functions
13 August, 2018

Check out a new paper from the Disney’s research team that introduces a modular convolutional architecture for denoising rendered images. The new method suggests mixing kernel-predicting networks with a number of task-specific modules and optimizing the assembly using an asymmetric loss. The team states that this new approach provides much better results. 

We present a modular convolutional architecture for denoising rendered images. We expand on the capabilities of kernel-predicting networks by combining them with a number of task-specific modules, and optimizing the assembly using an asymmetric loss. The source-aware encoder—the first module in the assembly—extracts low-level features and embeds them into a common feature space, enabling quick adaptation of a trained network to novel data. The spatial and temporal modules extract abstract, high-level features for kernel-based reconstruction, which is performed at three different spatial scales to reduce low-frequency artifacts. The complete network is trained using a class of asymmetric loss functions that are designed to preserve details and provide the user with a direct control over the variancebias trade-off during inference. We also propose an error-predicting module for inferring reconstruction error maps that can be used to drive adaptive sampling. Finally, we present a theoretical analysis of convergence rates of kernel-predicting architectures, shedding light on why kernel prediction performs better than synthesizing the colors directly, complementing the empirical evidence presented in this and previous works. We demonstrate that our networks attain results that compare favorably to state-of-the-art methods in terms of detail preservation, low-frequency noise removal, and temporal stability on a variety of production and academic datasets.


You can get more details by following this link (full paper) or attending this year’s SIGGRAPH. 

Leave a Reply

Be the First to Comment!