3DMeshNet
The underlying representation of 3D reconstruction algorithms has a major impact on the quality, memory size and robustness the reconstruction can achieve.
Goals:
- DNN which generates a triangle topology / wireframe from a depth image
- DNN which can merge/extend a already existing triangular scene mesh with new information from RGBD images (novel)
- DNN which can merge meshes to become manifolds (novel)
- DNN which can fill in holes or complete parts of the scene directly on the triangle mesh. (like https://github.com/angeladai/sgnn but based on triangles)