This is a repository for the paper "Improved Adversarial Systems for 3D Object Generation and Reconstruction". There are three GAN projects held here. The first, held in the folder 3D-Generation, is code for generating 3D objects. The second, held in the folder 3D-Translation, is code for producing 3D objects when conditioned on image inputs. The third, held in the folder 3D-reconstruction, is code for reconstructing 3D objects, from single perspective depth scans.
A diagram outlining the 3 Generative Adverserial Networks used in this repo.
Example 3D objects generated from a distribution constisting of 10 3D object classes in 12 orientations, rotated for easy viewing.
Comparison of 3D-IWGAN's generation ability compared to that of 3D-GAN's.