Denoising with Kernel Prediction and Asymmetric Loss Functions
Events
Subscribe:  iCal  |  Google Calendar
Los Angeles US   27, Sep — 1, Oct
Moscow RU   4, Oct — 8, Oct
Prague CZ   5, Oct — 7, Oct
São Paulo BR   10, Oct — 15, Oct
Latest comments

I feel like you can shoot this after printing it out...

by Oneiros
45 min ago

And yes, these video are made using the standard unity pipeline but we will release soon also the HDRP version. Stay tuned!

by Oneiros
50 min ago

Hi khiree.taylan, the projects could run in multiplatform. The graphics showed in these videos are made using a gtx 1070 but soon we will release also the mobile version :)

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

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.

Disney 

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!

avatar
wpDiscuz
Related articles