Smoothing Predictions of Multi-Task EmotiNet Models for Compound Facial Expression Recognition
Abstract
In this paper, we describe the results of our team that took the second place in the compound expression (CE) recognition task of the seventh Affective Behavior Analysis in-the-wild (ABAW) competition. We propose an efficient pipeline based on pre-trained frame-level emotional feature extractors (EmotiNet), e.g., MT-EmotiMobileFaceNet, to estimate valence-arousal and basic facial expressions given a facial photo. It was demonstrated that a significant step in improving the overall accuracy is the smoothing of neural network output scores using box filters. It was demonstrated that such a post-processing improves the F1-score of CE classification up to 5%.
Cite
Text
Savchenko. "Smoothing Predictions of Multi-Task EmotiNet Models for Compound Facial Expression Recognition." European Conference on Computer Vision Workshops, 2024. doi:10.1007/978-3-031-91581-9_18Markdown
[Savchenko. "Smoothing Predictions of Multi-Task EmotiNet Models for Compound Facial Expression Recognition." European Conference on Computer Vision Workshops, 2024.](https://mlanthology.org/eccvw/2024/savchenko2024eccvw-smoothing/) doi:10.1007/978-3-031-91581-9_18BibTeX
@inproceedings{savchenko2024eccvw-smoothing,
title = {{Smoothing Predictions of Multi-Task EmotiNet Models for Compound Facial Expression Recognition}},
author = {Savchenko, Andrey V.},
booktitle = {European Conference on Computer Vision Workshops},
year = {2024},
pages = {257-266},
doi = {10.1007/978-3-031-91581-9_18},
url = {https://mlanthology.org/eccvw/2024/savchenko2024eccvw-smoothing/}
}