FastSESR: Fast Scene-level Explicit Surface Reconstruction

Explicit surface reconstruction aims to recover high-fidelity meshes directly from point clouds. While existing methods achieve strong performance on scene-level data, they often rely on test-time optimization, resulting in a prohibitive runtime of several minutes. To address this bottleneck, we propose FastSESR, a two-stage framework for efficient scene-level explicit surface reconstruction. In the first stage, a lightweight triangular candidate network (TCN) captures local connections via an edge-factorized parameterization, enabling effective extraction of surface triangles from uniformly sampled points. In the second stage, an offset optimization network amortizes offset refinement into a small, fixed number of learnable update steps guided by TCN, producing geometries that are more suitable for triangulation. Experiments on multiple scene-level datasets show that FastSESR accelerates surface reconstruction by at least $20 \times$ over prior methods while maintaining competitive reconstruction quality. Moreover, evaluations on shape-level benchmarks indicate good generalization performance. Our code is available at https://github.com/Vaiduryasses/FastSESR.

3D point clouds are like digital scans made of many scattered points in space. To make these scans useful for robots, autonomous driving systems, or 3D modeling tools, we need to connect the points into a continuous triangle mesh surface. However, many high-quality reconstruction methods either require reliable surface normals, which are often difficult to obtain from raw scans, or rely on repeated optimization for every new scene, which can take several minutes. We propose FastSESR to make this process much faster. The method first learns which nearby points are likely to form surface triangles. It then uses a learned offset network to slightly adjust point positions so that clean and reasonable meshes can be extracted more easily. Instead of optimizing each scene again and again at test time, FastSESR performs the main reconstruction with a single forward pass. Experiments show that FastSESR reduces scene-level reconstruction time from minutes to seconds while maintaining competitive surface quality. This makes high-quality 3D scene reconstruction more practical for applications such as robotic perception, autonomous driving, and digital content creation.