This is the code and data for the paper of the above title. It consists of a large motion database containing several external databases retargetted to a single skeleton structure, as well as the code for the paper implemented in Theano.
Here is an overview of the structure:
`data` - The motion database and data processing scripts
`motion` - A helper library for dealing with motion data
`nn` - A custom Theano library for neural networks
`synth` - Code, examples, and demos used in the paper
Once you have installed Theano, Numpy, Scipy, as the network is already trained
and the data already preprocessed, you can start by running any of the demo
scripts in the synth directory.
This code and data is free for academic and non-commercial purposes but we would ask that you please include the following citations in any published work which uses this code or data.
@inproceedings{Holden:2015:LMM,
author = {Holden, Daniel and Saito, Jun and Komura, Taku and Joyce, Thomas},
title = {Learning Motion Manifolds with Convolutional Autoencoders},
booktitle = {SIGGRAPH Asia 2015 Technical Briefs},
year = {2015},
}
@inproceedings{Holden:2016:DLF,
author = {Holden, Daniel and Saito, Jun and Komura, Taku},
title = {A Deep Learning Framework for Character Motion Synthesis and Editing},
booktitle = {SIGGRAPH 2016},
year = {2016},
}
The motion databases have their own individual licenses that also ask for citation and have a number of conditions so please respect those usage terms.
If you have any questions please contact me at [email protected]