This is the official implementation of the paper STEP: Spatial Temporal Graph Convolutional Networks for Emotion Perception from Gaits. Please use the following citation if you find our work uesful:

@inproceedings{bhattacharya2020step,
author = {Bhattacharya, Uttaran and Mittal, Trisha and Chandra, Rohan and Randhavane, Tanmay and Bera, Aniket and Manocha, Dinesh},
title = {STEP: Spatial Temporal Graph Convolutional Networks for Emotion Perception from Gaits},
year = {2020},
publisher = {AAAI Press},
booktitle = {Proceedings of the Thirty-Fourth AAAI Conference on Artificial Intelligence},
pages = {1342–1350},
numpages = {9},
series = {AAAI’20}
}

We have also released the Emotion-Gait dataset with this code, which is available for download here: https://go.umd.edu/emotion-gait.

1. generator_cvae is the generator.

2. classifier_stgcn_real_only is the baseline classifier using only the real 342 gaits.

3. classifier_stgcn_real_and_synth is the baseline classifier using both real 342 and N synthetic gaits.

4. clasifier_hybrid is the hybrid classifier using both deep and physiologically-motivated features.

5. compute_aff_features consists of the set of scripts to compute the affective features from 16-joint pose sequences. Calling main.py with the correct data path computes the features, and save them in the affectiveFeatures<f_type>.h5 file, where f_type is the desired type of features:

   • '' original data (default)
   • 4DCVAEGCN data generated by the CVAE.