3D-Surface-Generation-with-Tangent-Convolutions

Title: 3D Surface Generation with Tangent Convolutions
Type of project: Semester Project
Year: 2019
Supervisor(s): Sandro Lombardi
Student: Parker Ewen

Setup

AtlasNet

Follow steps provided by original repo.

Tangent Convolutions

Build Open3D

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

Run Tangent Convolutions

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

References:

[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.

parkerewen5441 / 3d-surface-generation-with-tangent-convolutions Goto Github PK