Name: home-platform
Owner: HoME-Platform
Description: HoME: a Household Multimodal Environment is a platform for artificial agents to learn from vision, audio, semantics, physics, and interaction with objects and other agents, all within a realistic context.
Created: 2017-11-16 21:30:40.0
Updated: 2018-01-16 13:50:35.0
Pushed: 2018-01-07 22:29:12.0
Homepage: https://home-platform.github.io/
Size: 8302
Language: Python
GitHub Committers
| User | Most Recent Commit | # Commits |
|---|
Other Committers
| User | Most Recent Commit | # Commits |
|---|
HoME is a platform for artificial agents to learn from vision, audio, semantics, physics, and interaction with objects and other agents, all within a realistic context.
Check out the paper on Arxiv for more details: HoME: a Household Multimodal Environment
Main requirements:
To install dependencies on Ubuntu operating systems:
apt-get install python-pip python-tk python-dev build-essential libsndfile1 portaudio19-dev
pip2 install --upgrade pip numpy scipy matplotlib gym panda3d pysoundfile pyaudio resampy nose coverage Pillow
or, for Python 3:
apt-get install python3-pip python3-tk python3-dev build-essential libsndfile1 portaudio19-dev
pip3 install --upgrade pip numpy scipy matplotlib gym panda3d pysoundfile pyaudio resampy nose coverage Pillow
(Packages nose and coverage are for tests only and can be omitted)
Finally you have to install EVERT. In order to do so, please follow the instructions over at https://github.com/sbrodeur/evert
The Home environment is based on the SUNCG dataset.
Important! Before you can use this library, you need to obtain the SUNCG dataset. In order to do so, please follow the instructions on their website.
For the test suit we included a single small house as sample in this repository.
Download the source code from the git repository:
r -p $HOME/work
HOME/work
clone https://github.com/HoME-Platform/home-platform.git
Note that the library must be in the PYTHONPATH environment variable for Python to be able to find it:
rt PYTHONPATH=$HOME/work/home-platform:$PYTHONPATH
This can also be added at the end of the configuration file $HOME/.bashrc
To ensure all libraries where correctly installed, it is advised to run the test suite:
HOME/work/home-platform/tests
n_tests.sh
Note that this can take some time.
Before you start please read steps 0 and 1 so that you know what to expect. The data installation and preparation can take some time (because it's a big dataset).
The dataset can be acquired by contacting the authors at their website: http://suncg.cs.princeton.edu/
Currently the dataset downloaded consists of a single zip file, which in turn contains multiple zip files (one for each directory in [house, object, object_vox, room, texture]).
Please unzip all files so that as a result you have the folloging directory structure somewhere on your PC:
e/path/doesnt/matter/SUNCG/
e/path/doesnt/matter/SUNCG/house/
e/path/doesnt/matter/SUNCG/house/0004d52d1aeeb8ae6de39d6bd993e992/
e/path/doesnt/matter/SUNCG/house/0004dd3cb11e50530676f77b55262d38/
e/path/doesnt/matter/SUNCG/object/100/
e/path/doesnt/matter/SUNCG/object/101/
The unzipped files take up approx 28.1GB disk space on an NTFS-formatted drive.
Warning Unzipping may take considerable time and should not be done on a network drive
due to the overhead in network communication. We advise to extract everything on a local machine
and if necessary copy it to networked machines via rsync like so:
rsync -avh --info=progress2 --remove-source-files /some/path/doesnt/matter/SUNCG /cluster/datasets/
Once everything is unzipped, you can remove the original zip files.
As final step please “install” the dataset by (a) symlinking it into your home directory
ln -s /some/path/doesnt/matter/SUNCG/ ~/.suncg
or by (b) setting the environmental variable to the right path:
export SUNCG_DATA_DIR="/cluster/datasets/SUNCG"
If you do both of these things, the environmental variable takes precedent.
In order for Panda3D to be able to read the 3D files of the houses, you need to first convert them to a different format (from OBJ/Wavefront to EGG and BAM).
To do so, just cd into the scripts folder in this repository and run the conversion script:
# assuming you currently are in the directory of this readme file:
cd scripts
# if you installed the dataset via symlink:
./convert_suncg.sh ~/.suncg/ # the trailing slash is important
# if you installed the dataset via environmental vairable:
./convert_suncg.sh $SUNCG_DATA_DIR/ # the trailing slash might be important
Now sit back, have a tea, a dinner, maybe a little holiday somewhere warm - this will take a while.
TODO
(In the meantime please have a look at our tests in the tests directory.
There we cover pretty much all aspects of the library.)
TODO
(In the meantime please have a look at our tests in the tests directory.
There we cover pretty much all aspects of the library.)
TODO
(… you get the idea.)
TODO
(… you get the idea.)
TODO
(… you get the idea.)