TEXTRIX: Latent Attribute Grid for Native Texture Generation and Beyond

2025-12-02

AI Summary

Existing 3D texture generation methods suffer from inter-view inconsistency and incomplete coverage of complex surfaces due to multi-view fusion, compromising texture fidelity and geometric integrity. To address this, we propose TEXTRIX—a novel framework that introduces the first implicit 3D attribute grid, unifying texture synthesis and semantic segmentation directly in native voxel space and thereby eliminating the need for explicit view alignment. We further design a sparse-attention diffusion Transformer that jointly optimizes high-resolution texture generation and part-level semantic prediction on this grid. Our approach simultaneously enhances texture seamlessness, geometric consistency, and segmentation boundary accuracy. Extensive experiments demonstrate state-of-the-art performance on both 3D texture generation and 3D semantic segmentation benchmarks.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Prevailing 3D texture generation methods, which often rely on multi-view fusion, are frequently hindered by inter-view inconsistencies and incomplete coverage of complex surfaces, limiting the fidelity and completeness of the generated content. To overcome these challenges, we introduce TEXTRIX, a native 3D attribute generation framework for high-fidelity texture synthesis and downstream applications such as precise 3D part segmentation. Our approach constructs a latent 3D attribute grid and leverages a Diffusion Transformer equipped with sparse attention, enabling direct coloring of 3D models in volumetric space and fundamentally avoiding the limitations of multi-view fusion. Built upon this native representation, the framework naturally extends to high-precision 3D segmentation by training the same architecture to predict semantic attributes on the grid. Extensive experiments demonstrate state-of-the-art performance on both tasks, producing seamless, high-fidelity textures and accurate 3D part segmentation with precise boundaries.

Problem

• Generates high-fidelity 3D textures directly in volumetric space
• Overcomes multi-view fusion inconsistencies and incomplete surface coverage
• Extends native 3D attribute generation to precise part segmentation

Innovation

• Latent 3D attribute grid for direct volumetric coloring
• Diffusion Transformer with sparse attention mechanism
• Unified architecture for texture synthesis and segmentation