ExMesh: EXplicit Mesh Reconstruction with Topology Adaptation

2026-06-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

ExMesh is a novel framework for reconstructing high-fidelity surface meshes directly from multi-view images, addressing challenges in adaptive topology refinement and consistent UV coordinate maintenance. Developed by the University of Science and Technology of China, ExMesh integrates differentiable optimization with discrete topology updates, featuring an adaptive vertex splitting and merging strategy and real-time UV maintenance. This approach enables coarse-to-fine optimization while preserving geometric integrity and decoupling texture resolution from face count. Experiments on the DTU and NeRF-synthetic datasets, using a single NVIDIA RTX 3090 GPU, demonstrate ExMesh achieves comparable geometric accuracy to state-of-the-art methods like Neuralangelo and Nvdiffrec, but with significantly fewer faces and shorter training times, balancing accuracy, efficiency, and mesh conciseness.

Key takeaway

For 3D graphics developers or ML engineers focused on high-fidelity 3D asset generation, ExMesh provides a robust solution for direct mesh reconstruction. You can leverage its adaptive topology and decoupled UV mapping to create structurally clean, editable meshes with detailed textures, avoiding the artifacts and complexity of intermediate representations. This framework offers a superior balance of accuracy and efficiency, making it ideal for applications requiring real-time rendering and direct scene editing.

Key insights

ExMesh directly optimizes explicit meshes by integrating differentiable optimization with adaptive topology and real-time UV maintenance.

Principles

• Direct explicit mesh optimization avoids intermediate representations and post-processing artifacts.
• Adaptive topology refinement improves detail recovery in complex regions and removes redundant faces.
• Decoupled UV maps enable high-fidelity textures on concise meshes, independent of face count.

Method

ExMesh iteratively optimizes geometry (vertex positions) and a separate UV map using a differentiable renderer, interleaved with adaptive vertex splitting (based on gradient/curvature) and merging (based on visibility/degeneracy) for coarse-to-fine refinement.

In practice

• Reconstruct high-fidelity, editable 3D meshes from multi-view images.
• Achieve efficient real-time rendering with concise mesh structures.
• Maintain texture detail independently of mesh face count.

Topics

• Mesh Reconstruction
• Differentiable Rendering
• Topology Adaptation
• UV Mapping
• Multi-view Stereo
• Explicit Meshes

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• ExMesh: EXplicit Mesh Reconstruction with Topology Adaptation — ExMesh directly optimizes explicit meshes by combining differentiable optimization with adaptive discrete topology updates and real-time UV maintenance.
• City-Mesh3R: Simulation-Ready City-Scale 3D Mesh Reconstruction from Multi-View Images — City-Mesh3R reconstructs simulation-ready, watertight city-scale 3D meshes from multi-view images using a scalable, end-to-end divide-and-conquer approach.
• Pixel Cube: Diffusion-based Portrait Video Relighting Through Realistic Lighting Reproduction — Diffusion models, trained on hybrid real/synthetic data from a specialized LED stage, enable photorealistic, temporally consistent portrait video relighting.
• EvoGS: Constructing Continuous-Layered Gaussian Splatting with Evolution Tree for Scalable 3D Streaming — EvoGS uses a continuous, wavelet-inspired parent-child refinement tree for scalable 3DGS, eliminating redundancy and improving streaming.
• SQuadGen: Generating Simple Quad Layouts via Chart Distance Fields — SQuadGen uses diffusion models and Chart Distance Fields to generate simple, artist-friendly quad layouts for 3D shapes.
• Real2SAM2Real: Generative 3D Caches as Complementary Context for Video Diffusion — Real2SAM2Real uses editable 3D caches from lifting models to provide robust geometric scaffolds for Video Diffusion Models, improving control and consistency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.