Skip to main content
GameDev.net gamedev.net

Neural Networks News

The latest Neural Networks coverage curated for game developers.

LayoutShop: Content-Constrained Exploratory Design of Creative Article Layout

A new layout-generation framework, LayoutShop, uses content constraints and optimization to build creative article compositions from templates. It extracts eligible structures, solves their geometry, then ranks results with neural assessment …

Research arXiv cs.GR · 2 weeks, 2 days ago
Proximity-Preserving Neural Subdivision

A new neural subdivision scheme aims to keep the benefits of classic Loop refinement while learning local detail. Proximity-Preserving Neural Subdivision adds a bounded, curvature-gated correction that stays rigid-motion equivariant …

Research arXiv cs.GR · 1 month ago
Ragged Neighborhood Attention for Spatiotemporal Neural Denoising of Deep Monte Carlo Renderings

Disney Research Studios has introduced RaNA, a ragged neighborhood attention operator for denoising deep Monte Carlo renders. The method powers RaNAD, the first spatiotemporal neural denoiser for deep images, and …

Research Disney Research Studios · 2 months, 1 week ago
SCALMU: Synthetically-trained Coupling of Adaptive Learned Multiplicative Updates for Hyperspectral-Multispectral Fusion

A new hyperspectral-multispectral fusion model, SCALMU, blends classical CNMF structure with learned multiplicative updates to recover high-res hyperspectral images. The big practical angle for graphics and imaging teams is that …

Research arXiv cs.GR · 3 months, 3 weeks ago
PureSample: Neural Materials Learned by Sampling Microgeometry

PureSample introduces a novel neural BRDF representation that simplifies the process of material appearance learning through microgeometry sampling. This advancement is particularly relevant for graphics programmers, as it streamlines BRDF …

Research arXiv cs.GR · 4 months, 2 weeks ago
SAND: Spatially Adaptive Network Depth for Fast Sampling of Neural Implicit Surfaces

The introduction of SAND, a spatially adaptive network depth framework, significantly enhances the efficiency of neural implicit surface sampling. This innovation allows graphics programmers to optimize computational resources by adjusting …

Research arXiv cs.GR · 4 months, 3 weeks ago
GMT: A Geometric Multigrid Transformer Solver for Microstructure Homogenization

The introduction of GMT, a Geometric Multigrid Transformer, significantly accelerates lattice homogenization processes, achieving a remarkable 160x speedup. This advancement is particularly relevant for graphics programmers and designers working with …

Research arXiv cs.GR · 4 months, 3 weeks ago
8DNA: 8D Neural Asset Light Transport by Distribution Learning

The introduction of 8D Neural Asset Light Transport (8DNA) marks a significant advancement in rendering technology. By pre-baking complex light transport effects into neural representations, this method allows for accurate …

Research arXiv cs.GR · 4 months, 3 weeks ago
Neural Enhancement of Analytical Appearance Models

A novel framework called neural enhancement is introduced, combining traditional analytical models with neural networks to improve graphics rendering. This approach allows for more accurate and efficient representation of materials, …

Research arXiv cs.GR · 4 months, 3 weeks ago
Voxel Deformation-Aware Neural Intersection Function

The introduction of the Voxel Deformation-Aware Neural Intersection Function (VDNIF) is a significant advancement for graphics programmers. This method enhances the Locally-Subdivided Neural Intersection Function by enabling accurate intersection predictions …

Research arXiv cs.GR · 4 months, 3 weeks ago
PEPS: Positional Encoding Projected Sampling -- Extended

The Positional Encoding Projected Sampling (PEPS) method introduces a significant advancement in implicit neural representations, particularly for graphics programmers. By reducing parameter requirements by 25% while maintaining reconstruction quality, this …

Research arXiv cs.GR · 4 months, 3 weeks ago
Quasi-Medial Distance Field (Q-MDF): A Robust Method for Approximating and Discretizing Neural Medial Axes

The introduction of the Quasi-Medial Distance Field (Q-MDF) offers a significant advancement for graphics programmers and artists. This method enhances the approximation and discretization of neural medial axes, enabling more …

Research ACM Graphics · 4 months, 4 weeks ago
Neural Gabor Splatting: Enhanced Gaussian Splatting with Neural Gabor for High-frequency Surface Reconstruction

Neural Gabor Splatting introduces a significant advancement in 3D surface reconstruction by enhancing Gaussian splatting techniques. This method allows for better handling of high-frequency details, which is crucial for graphics …

Research arXiv cs.GR · 5 months ago
Neural Dynamic GI: Random-Access Neural Compression for Temporal Lightmaps in Dynamic Lighting Environments

Neural Dynamic GI introduces a groundbreaking compression technique for temporal lightmaps, significantly reducing storage needs while maintaining high-quality global illumination. This innovation is particularly relevant for graphics programmers and artists …

Research arXiv cs.GR · 5 months, 1 week ago
Cooperative Vectors Introduction

GPU neural-network work is moving past pure inference: Evolve’s renderer team is using Cooperative Vectors to make divergent, per-pixel NN evaluation practical in shaders. The big win is avoiding awkward …

Blog Evolve Blog · 5 months, 1 week ago
NeuVolEx: Implicit Neural Features for Volume Exploration

NeuVolEx introduces a novel approach to volume exploration using implicit neural representations (INRs). This advancement is particularly relevant for graphics programmers and designers, as it enhances the classification and clustering …

Research arXiv cs.GR · 5 months, 1 week ago
Fatigue-PINN: Physics-Informed Fatigue-Driven Motion Modulation and Synthesis

Fatigue-PINN introduces a novel approach to motion synthesis by integrating physics-informed neural networks to model human fatigue. This advancement allows for smoother animations and more realistic character movements, which is …

Research arXiv cs.GR · 5 months, 1 week ago
Physics-Based Motion Tracking of Contact-Rich Interacting Characters

A new method for physics-based motion tracking addresses the challenges of contact-rich interactions in character animation. This approach, utilizing a progressive neural network, enhances stability and efficiency in motion tracking, …

Research arXiv cs.GR · 5 months, 1 week ago
Real-time Neural Six-way Lightmaps

The introduction of neural six-way lightmaps marks a significant advancement in real-time rendering for games. This technique allows for dynamic smoke effects that respond to camera movement and lighting changes, …

Research arXiv cs.GR · 5 months, 2 weeks ago
CANRIG: Cross-Attention Neural Face Rigging with Variable Local Control

CANRIG introduces a groundbreaking approach to facial rigging, significantly reducing the time and effort required for animators. By leveraging neural networks, it allows for both global and local control of …

Research Disney Research Studios · 5 months, 2 weeks ago
Page 1 of 3 Next