Title: 3D Surface Generation with Tangent Convolutions
Type of project: Semester Project
Year: 2019
Supervisor(s): Sandro Lombardi
Student: Parker Ewen
Follow steps provided by original repo.
Following steps were provided by original repo with some extra steps:
Make sure venv is using Python 3.6
$ python -V
$ python3 -V #depending on venv
Clone this version of Open3D
$ cd Open3D/src/Python/
$ mkdir Helper
$ mkdir Test
$ cd ../../
$ util/scripts/install-deps-ubuntu.sh
$ mkdir build
$ cd build
$ cmake ../src
$ make
To test whether Open3D was built, test using the following:
$ cd Open3D/build/lib/
$ python
$ from py3d import *
There may be issues with building Open3D. If that is the case, check the following:
- Open CMakeCache.txt in Open3D/build and make sure PYTHON_EXECUTABLE:FILEPATH and PYTHON_LIBRARY:FILEPATH= point to python3.6 and python3.6m.so respectively
- Python 3.6 isn't supported by Ubuntu 16.04. You need to recreate the virtual environment after you install the python3.6-dev package
$ sudo add-apt-repository ppa:jonathonf/python-3.6
$ sudo apt update
$ sudo apt install python3.6 python3.6-dev
$ sudo apt remove python3-dev # to ensure the system's Python 3.5 wouldn't interfere
### Create new virtualenv with flag --python=/usr/bin/python3.6
# In new virtualenv...
$ pip install dlib
Download one of the datasets specified by their GitHub page.
As an example:
$ python get_data.py <directory_where_zip_is> /path/to/tangent_conv/data/raw/stanford stanford
Follow their procedure, starting with precomputations.
<experiment_config> refers to the .json config path. As an example:
$ python tc.py experiments/stanford/dhnrgb/config.json --precompute
[1] Groueix, Thibault, et al. "A papier-mâché approach to learning 3d surface generation." Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.
[2] Tatarchenko, Maxim, et al. "Tangent convolutions for dense prediction in 3d." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018.