MeshLRM: Large Reconstruction Model for High-Quality Mesh
Approach for reconstructing high-quality meshes from four input images in less than one second, combining differentiable mesh extraction and rendering within LRM
MeshLRM presents a Large Reconstruction Model designed for high-quality mesh reconstruction. The model consists of two stages: MeshLRM (NeRF) for initial reconstruction using NeRF and MeshLRM for surface rendering fine-tuning. The first stage, MeshLRM (NeRF), utilizes a large-scale NeRF model to reconstruct high-quality 3D meshes from sparse views. The second stage refines the mesh using surface rendering techniques, including ray opacity and normal losses to improve geometry accuracy.
Comments
None