ShapeCompletion3D / shape_completion

Top level repo for shape complete project

Shape Completion

Top level repo for shape complete project.

Looking for shape_completion_experiments? It's now here: https://github.com/CURG/shape_completion_experiments

Dependencies

Should be manually installed before setup:

• Theano
• H5py
• ROS

Automatically installed by wstool during setup:

• See the .rosinstall file for a list of these.

Setup

The following is the procedure to set up the workspace for the Shape Completion project.

1. Install the dependencies mentioned above under the subsection "Should be manually installed before setup."
2. Clone this repository into your workspace.

git clone https://github.com/ShapeCompletion3D/shape_completion.git
cd shape_completion/

3. Run wstool update. This will clone all the other dependent repositories listed in the .rosinstall file into that workspace

wstool update

4. Run the installation script ( install.sh ) from the repository. The script will set up all the dependent repositories

bash install.sh

5. On your home directory create a .theano_rc file with the following contents:

[global]
floatX = float32
device = gpu

6. On your home directory add the following lines to your .bash_rc file:

export CUDA_HOME=/usr/local/cuda-6.5
export LD_LIBRARY_PATH=${CUDA_HOME}/lib64:$LD_LIBRARY_PATH
export PATH=${CUDA_HOME}/bin:${PATH}

7. Start a new terminal window so that the .bash_rc changes take effect.
8. Verify that all the datasets we have, including the YCB dataset, is stored in /srv/data/

Running an experiment

1. Run set_paths.sh to add the necessary directories to your PYTHONPATH.

source set_paths.sh

2. Go to the experiments folder inside the shape_completion_experiments repository

cd train/shape_completion_experiments/experiments/

3. Run an experiment script. For example, reconstruction_3d_ycb.py, which is the shape completion experiment that uses the YCB dataset with an input and output size of 30x30x30:

python reconstruction_3d_ycb.py