Unsupervised Learning of Object Structure and Dynamics from Videos

This reprository contains code for the model described in https://arxiv.org/abs/1906.07889.

Also see the project website.

Usage

Installation

This code has been tested on Linux.

1. Clone the repository.

git clone -b master --single-branch https://github.com/google-research/google-research.git

2. Make sure google-research is the current directory:

cd google-research

3. Create and activate a new virtualenv:

virtualenv -p python3 video_structure
source video_structure/bin/activate

4. Install required packages:

pip install -r video_structure/requirements.txt

Run minimal example

Run on GPU device 0:

CUDA_VISIBLE_DEVICES=0 python -m video_structure.train

Run on CPU:

CUDA_VISIBLE_DEVICES= python -m video_structure.train

Citation

@inproceedings{minderer2019unsupervised,
	title = {Unsupervised Learning of Object Structure and Dynamics from Videos},
	author = {Minderer, Matthias and Sun, Chen and Villegas, Ruben and Cole, Forrester and Murphy, Kevin and Lee, Honglak},
	booktitle = {arXiv: 1906.07889},
	year = {2019},
}