GATE: Geometry-Aware Trained Encoding

2025-06-09

AI Summary

Existing hash-based encodings in neural rendering suffer from hash collisions, coupling between resolution and scene scale, and irregular memory access patterns. To address these issues, this paper proposes GATE (Geometric-Aware Triangular Embedding), a novel encoding method that embeds trainable feature vectors onto triangular mesh surfaces—marking the first geometric-aware parametric encoding for neural rendering. GATE decouples feature density from geometric density to enable adaptive level-of-detail control; employs mesh shading to dynamically modulate local feature density; and integrates with neural radiance caching to regularize memory access. Experiments demonstrate that GATE eliminates encoding collisions while maintaining low memory overhead, significantly improving geometric consistency and training stability. In multi-scale neural radiance field reconstruction, GATE achieves superior reconstruction accuracy and enhanced convergence controllability compared to state-of-the-art hash-based approaches.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

The encoding of input parameters is one of the fundamental building blocks of neural network algorithms. Its goal is to map the input data to a higher-dimensional space, typically supported by trained feature vectors. The mapping is crucial for the efficiency and approximation quality of neural networks. We propose a novel geometry-aware encoding called GATE that stores feature vectors on the surface of triangular meshes. Our encoding is suitable for neural rendering-related algorithms, for example, neural radiance caching. It also avoids limitations of previous hash-based encoding schemes, such as hash collisions, selection of resolution versus scene size, and divergent memory access. Our approach decouples feature vector density from geometry density using mesh colors, while allowing for finer control over neural network training and adaptive level-of-detail.

Problem

• Proposes geometry-aware encoding for neural networks
• Addresses limitations of hash-based encoding schemes
• Enhances neural rendering with adaptive level-of-detail

Innovation

• Geometry-aware encoding stores features on meshes
• Avoids hash collisions and divergent memory access
• Decouples feature density using mesh colors