ExMesh: Explicit Mesh Reconstruction with Topology Adaptation

ExMesh is a new mesh reconstruction framework built to optimize explicit meshes directly from multi-view images, instead of relying on intermediate volumes and cleanup passes such as Marching Cubes or TSDF fusion. The goal is straightforward: produce cleaner geometry with fewer artifacts and less fragmentation, while keeping the pipeline efficient enough to matter in production-adjacent workflows.

The technical wrinkle is topology. As meshes gain detail, they need to split and merge adaptively without collapsing into degenerate faces or breaking texture coordinates. ExMesh addresses that by pairing differentiable optimization with discrete topology operations, including adaptive vertex splitting and merging plus real-time UV maintenance. That combination is what makes the system interesting for game-dev tooling, since geometry quality and texturing stability usually fight each other during reconstruction.

The authors position ExMesh as the first framework to integrate discrete topology changes into a continuous differentiable optimization pipeline. In practice, that means coarse-to-fine refinement can happen while preserving geometric integrity, rather than forcing a later cleanup stage to guess what the optimizer intended. The stated tradeoff is a balance among accuracy, computational efficiency, and mesh conciseness.

For developers, the relevance is strongest anywhere reconstructed meshes feed directly into asset pipelines: photogrammetry, scan cleanup, virtual production, and automated content generation. If the approach holds up outside the paper, it could reduce...