The Multitask EmotionNet in the ABAW3 challenge

We participated in the ABAW3 MTL Challenge held in conjunction with CVPR2022. Our team name is NISL-2022.

Leaderboard is here. We won the first place in the MTL challenge!

This repository contains the code for our Multitask EmotionNet (the static and temporal approaches).

Dependency

Install the dependencies with the requirements.txt in MTL/:

pip install requirements.txt

Model Architecture

Pretrained Model

The pretrained model can be downloaded from this link.

This model was trained on the training set of the Aff-wild2 dataset, and evaluated on the validation set of the Aff-wild2 dataset.

The validation metrics are listed below:

F1-AU	F1-EXPR	CCC-V	CCC-A
0.548	0.518	0.447	0.499

For more details about this approach, please refer to our Axiv paper.