Dynamic Emotion Modeling with Learnable Graphs and Graph Inception Network

Jan 11, 2021

• First release of the project.

Figure: Qualitative results showing the node (frame) for a graph input that generated the strongest response in our network

In this project, we present the Learnable Graph Inception Network (L-GrIN) that jointly learns to recognize emotion and to identify the underlying graph structure in the dynamic data. Our architecture comprises multiple novel components: a new graph convolution operation, a graph inception layer, learnable adjacency, and a learnable pooling function that yields a graph-level embedding. We evaluate the proposed architecture on five benchmark emotion recognition databases spanning three different modalities (video, audio, motion capture), where each database captures one of the following emotional cues: facial expressions, speech and body gestures.

Dependency installation

The code was successfully built and run with these versions:

pytorch-gpu 1.2.0
cudnn 7.6.4
cudatoolkit 10.0.130
opencv 3.4.2
scikit-learn 0.21.2
face_alignment 1.0.1 (for preprocessing)

Note: You can also create the environment I've tested with by importing environment.yml to conda.

Preprocessing Data

The process for RML database is in preprocess directory. The process converts the database into one txt file including graph structure and node attributes.

Note: you can download the processed data from here and put in this directory:

/dataset/
  Mine_Graph_RML/
    Mine_Graph_RML.txt

Training

You can train a simple model with main_Inception or the full model with main_Inception_learning_adj_pool

usage: main_Inception.py

optional arguments:
  -h, --help          Show this help message and exit
  -device             Which gpu to use if any
  -batch_size         Input batch size for training
  -iters_per_epoch    Number of iterations per each epoch
  -epochs             Number of epochs to train
  -lr                 Learning rate
  --num_layers        Number of inception layers
  --num_mlp_layers    Number of layers for MLP in each inception layer
  --hidden_dim        Number of hidden units for MLP
  --final_dropout     Dropout for classifier layer
  --Normalize         Normalizing data
  --patience          Patience for early stopping

Reference

If you found this repo useful give me a star!

ArXiv's paper

@article{shirian2020learnable,
  title={Learnable Graph Inception Network for Emotion Recognition},
  author={Shirian, Amir and Tripathi, Subarna and Guha, Tanaya},
  journal={arXiv preprint arXiv:2008.02661},
  year={2020}
}