NVIDIA Research Presenting 20 Papers at NeurIPS 2021

At the forefront of AI innovation, NVIDIA continues to push the boundaries of technology in machine learning, self-driving cars, robotics, graphics, and more.

At the forefront of AI innovation, NVIDIA continues to push the boundaries of technology in machine learning, self-driving cars, robotics, graphics, and more. NVIDIA researchers will present 20 papers at the thirty-fifth annual conference on Neural Information Processing Systems (NeurIPS) from December 6 to December 14, 2021. 

Here are some of the featured papers:

Alias-Free Generative Adversarial Networks (StyleGAN3)
Tero Karras, Miika Aittala, Samuli Laine, Erik Härkönen, Janne Hellsten, Jaakko Lehtinen, Timo Aila | Paper  | GitHub | Blog

StyleGAN3, a model developed by NVIDIA Research, will be presented on Tuesday, December 7 from 12:40 AM – 12:55 AM PST, advances the state-of-the-art in generative adversarial networks used to synthesize realistic images. The breakthrough brings graphics principles in signal processing and image processing to GANs to avoid aliasing: a kind of image corruption often visible when images are rotated, scaled or translated.

Video 1. Results from the StyleGAN3 model

EditGAN: High-Precision Semantic Image Editing
Huan Ling*, Karsten Kreis*, Daiqing Li, Seung Wook Kim, Antonio Torralba, Sanja Fidler | Paper | GitHub

EditGAN, a novel method for high quality, high precision semantic image editing, allowing users to edit images by modifying their highly detailed part segmentation masks, e.g., drawing a new mask for the headlight of a car. EditGAN builds on a GAN framework that jointly models images and their semantic segmentations, requiring only a handful of labeled examples, making it a scalable tool for editing. The poster session will be held on Thursday, December 9 from 8:30 AM – 10:00 AM PST.

Video 2. The video showcases EditGAN in an interactive demo tool.

SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers
Enze Xie, Wenhai Wang, Zhiding Yu, Anima Anandkumar, Jose M. Alvarez, Ping Luo | Paper | GitHub

SegFormer, a simple, efficient yet powerful semantic segmentation framework which unifies Transformers with lightweight multilayer perception (MLP) decoders. SegFormer has two appealing features: 1) SegFormer comprises a novel hierarchically structured Transformer encoder which outputs multiscale features. It does not need positional encoding, thereby avoiding the interpolation of positional codes which leads to decreased performance when the testing resolution differs from training. 2) SegFormer avoids complex decoders. The poster will be presented on Tuesday, December 7 from 8:30 AM – 10:00 AM PST.

Video 3. The video shows the excellent zero-shot robustness of SegFormer on the Cityscapes-C dataset.

DIB-R++: Learning to Predict Lighting and Material with a Hybrid Differentiable Renderer
Wenzheng Chen, Joey Litalien, Jun Gao, Zian Wang, Clement Fuji Tsang, Sameh Khamis, Or Litany, Sanja Fidler | Paper

DIB-R++, a deferred, image-based renderer which supports these photorealistic effects by combining rasterization and ray-tracing, taking advantage of their respective strengths—speed and realism.

Image 1. DIB-R++ is a hybrid renderer that combines rasterization and ray tracing together. Given a 3D mesh M, we employ (a) a rasterization-based renderer to obtain diffuse albedo, surface normals and mask maps. In the shading pass (b), we then use these buffers to compute the incident radiance by sampling or by representing lighting and the specular BRDF using a spherical Gaussian basis. Depending on the representation used in (c), we can render with advanced lighting and material effect (d).

In addition to the papers at NeurIPS 2021, researchers and developers can accelerate 3D deep learning research with new Kaolin features:

Kaolin is launching new features to accelerate 3D deep learning research. Updates to the NVIDIA Omniverse Kaolin app will bring robust visualization of massive point clouds. Updates to the Kaolin library will include support for tetrahedral meshes, rays management functionality, and a strong speedup to DIB-R. To learn more about Kaolin, watch the recent GTC session.

Image 2. Results from NVIDIA Kaolin

To view the complete list of NVIDIA Research accepted papers, workshop and tutorials, demos, and to explore job opportunities at NVIDIA, visit the NVIDIA at NeurIPS 2021 website.

Source:: NVIDIA