1suancaiyu / take_an_emotion_walk

Take an Emotion Walk: Perceiving Emotions from Gaits Using Hierarchical Attention Pooling and Affective Mapping

Take an Emotion Walk

This is the official implementation of the paper Take an Emotion Walk: Perceiving Emotions from Gaits Using Hierarchical Attention Pooling and Affective Mapping. Please add the following citation in your work if you use our code:

@InProceedings{bhattacharya2020taew,
author="Bhattacharya, Uttaran
and Roncal, Christian
and Mittal, Trisha
and Chandra, Rohan
and Kapsaskis, Kyra
and Gray, Kurt
and Bera, Aniket
and Manocha, Dinesh",
editor="Vedaldi, Andrea
and Bischof, Horst
and Brox, Thomas
and Frahm, Jan-Michael",
title="Take an Emotion Walk: Perceiving Emotions from Gaits Using Hierarchical Attention Pooling and Affective Mapping",
booktitle="Computer Vision -- ECCV 2020",
year="2020",
publisher="Springer International Publishing",
address="Cham",
pages="145--163",
isbn="978-3-030-58607-2"
}

Installation Requirements

Our scripts have been tested on Ubuntu 18.04 LTS with

• Python 3.6
• Cuda 10.2
• cudNN 7.6.5

We recommend using an Anaconda virtual environment. If Anaconda is not already installed, Install Anaconda and run

conda env create -n taew -f environment.yml

from within the project directory

Download datasets and network weights

Run the following command from within the project directory to download and extract the sample datasets and network weights:

sh download_data_weights.sh

We have used the Emotion-Gait dataset for this work. The full dataset is available for download here: https://go.umd.edu/emotion-gait.

Evaluation

1. Activate the conda environment

conda activate taew

2. Run the evaluation script For dgnn evaluation:

python evaluate.py --dgnn 

For stgcn evaluation:

python evaluate.py --stgcn

For lstm network evaluation:

python evaluate.py --lstm

For step evaluation:

python evaluate.py --step

For taew evaluation:

python evaluate.py --taew

Details for using taew_net as stand-alone

1. main.pyis the starting point of the code. It is runnable out-of-the-box once the datasets directory is downloaded and extracted. It also contains the full list of arguments for using the code.
2. utils/loader.py is used for loading the data and the labels. Labels are only available for the annotated part of the data.
3. utils/processor.pycontains the main training routine with forward and backward passes on the network, and parameter updates per iteration.
4. net/hapy.py contains the overall network and description of the forward pass.