Human Motion Embedding

Dependencies

• Tensorflow 1.2
• Keras 2.1
• Tqdm (progress bar)

Quickstart

1. Clone this repo and download the dataset in exponential map

git clone https://github.com/ytixu/Human_Motion_Embedding.git
cd Human_Motion_Embedding
mkdir data
cd data
wget http://www.cs.stanford.edu/people/ashesh/h3.6m.zip
unzip h3.6m.zip
rm h3.6m.zip
cd ..

2. Preprocess data

• Downsample from 50fps to 25 fps
• Convert to euler angle or cartesian coordinates, or keep in exponential map

cd src/utils
python data_preprocessing.py

Default conversion to euler angle. See input options for other parameterizations and for visualization of the motions.

3. Run

python train.py -m VL_RNN

4. Plot results

python viz_training_curves.py -p ../../out/vl_rnn/<OUTPUT LOG FILE NAME>