3D Mesh Reconstruction
Overview
3D mesh reconstruction estimates a connected surface from inputs such as single or multiple images, video, depth scans, LiDAR, and medical volumes. Methods include geometric fusion and surface fitting, parametric templates for people and faces, and neural approaches that extract meshes from implicit fields or 3D Gaussian splats; newer feed-forward and generative models aim to handle sparse views and large scenes quickly. Evaluations commonly use geometric measures such as Chamfer distance and F-score alongside rendering quality, topology, and performance on tasks such as grasping or physical simulation, with datasets including DTU and Tanks and Temples. A central challenge is that a visually convincing rendering may hide inaccurate, incomplete, or non-watertight geometry, so balancing appearance, surface fidelity, physical usability, and reliable evaluation remains an open tension.
Latest papers
109 papers13 papers in the last four weeks, up 86% on the four weeks before. 0.1% of all new papers.