A project for visual 3D reconstruction, where an automatic pipeline models, incrementally, the scene acquired from the images. The pipeline adopts a manifold reconstruction from Structure from Motion point cloud method, a novel mesh merging algorithm and photometric surface evolution.
Subsequent evolution involves improvements to the Multi-View Stereo pipeline, from depth map estimation to mesh refinement by combining classical approaches, such as PatchMatch, with machine learning oriented method, such as geometric deep learning.
Contact: Matteo Matteucci