This is the directory for recovering 3D object shape from single perspective depth scans using my method described in: https://arxiv.org/abs/1707.09557 .

This system is set up to only run on recovering chairs, though it is easy to look into the code to alter this. To set up the dataset call ./3D_generation.sh. This will download the shapenet models needed, and produce the chair surfaces, look into Kinect_setup.py to sort out the details for this. To run the model on this data call Kinect-VAE-3D-IWGAN.py, there are other parameters which can be specified, call with -h to view them. To evaluate the output you can use the visualize.py script. To use this just call python visualize.py VOXEL_FILE, where voxel file is the file created during training and saved to savepoint/NAME/EPOCH.npy. Graphs will be created which allow you to track the discriminator's loss, which will at first raise rapidly but then begin to decrease, and this decreasing loss can be used to track convergence. 
Let me know if you have any issues at [email protected]