Project of the MVA course "Object Recognition and Computer Vision". In this project we study the following methods to generate 3D meshes from a single 2D point of view of an object:
- AtlasNet
- Pixel2Mesh After reproducing the results of both papers on a single Shapenet category (cars), we provided some quantitavive and qualitative comparisons. Finally, we adapted mesh regularization schemes presented in Pixel2Mesh on AtlasNet.
The code we used was taken from the methods' official GitHub pages (with slight modifications):
The car subset of Shapenet used is available here.
- Mathieu Orhan - MVA Student
- Bastien Dechamps - MVA Student