thibaultgroueix / hsp Goto Github PK

Install torch

Download CImg and place it in the torch-hsp subfolder. The file "CImg.h" needs to be in the path "torch-hsp/CImg/".

Install the torch package torch-hsp by running "luarocks make hsp-1.0-0.rockspec" in the torch-hsp folder

A demo script is included which reconstructs a single image and outputs a mesh as obj file. It needs as input the pretrained network provided here.

th hspDemo.lua <GPU ID> <Trained Network File Name> <Input Image File Name>

Example parameter files are provided here.

The data is provided here.

To train a network the paths to the shapenet dataset and the output folder in the "parameters.lua" file need to be adjusted first.

th trainNetworkHierarchical.lua <GPU ID> <Parameter File Name>

The code is released as GPLv2.

When using the provided data make sure to respect the shapenet license.

Please cite our paper when using the code.
C. Häne, S. Tulsiani, J. Malik, Hierarchical Surface Prediction for 3D Object Reconstruction, Proc. Int. Conf. on 3D Vision (3DV), 2017