SSRNet

Setup

Pre-prequisites for python project

python == 3.6
tensorflow >=1.3
joblib
trimesh

For C++ project

cudpp
lz4
flann_cuda

Build this version of Open3D(Rewritten by us)

$ cd Open3D
$ mkdir build
$ cd build
$ cmake ../src
$ make

Update the path to Open3D in python project

ssrnet/util/path_config.py

Experiments

Experimental parameters are stored in .json configuration files. We provide the prepared data for you:

ground truth1: https://pan.baidu.com/s/1dfTZstJSs173kDNjdfDXcw password: 7jkm

ground truth2: https://pan.baidu.com/s/1Um1A6DagiFeOrBjTc7Grzw password: ggb8

The transformed data consists of vertex labels, dividing information, and other necessary information. You can use it to run our network directly.

You can put the data anywhere, but make sure the data path in your configuration file are set correctly.

Training

To start network training, run

$ python ssr.py <config> --train

Generate mesh

First, test a trained model to predict vertex labels.Run

$ python ssr.py <config> --test

Second, use the output label file to generate mesh.Run

$ mesh_generator <scan_path> <label_file> <output_mesh_file> 

Example:

$ mesh_generator .../scan001/ .../pre_label.txt .../output/