Fine-tuning Technologies for Reducing the FER Bias Across Various Distributions

DOI

10.2991/978-94-6463-540-9_92How to use a DOI?

Keywords

Computer Version; Facial Expression Recognition; Transfer Learning

Abstract

Lacking sufficient data has become a serious problem in the field of Facial Expression Recognition (FER), since the cost of collecting a large amount of facial expression images is huge and training a new FER model from the beginning is time-consuming. In this paper, the author trained a FER model based on a gray-scale dataset (FER2013) and found several shortages in both the dataset and the model. In order to achieve better accuracy and reduce the bias in the previous training domain, the author searched for a new dataset and applied transfer learning to transfer the FER model to the new domain. More specifically, this study was based on the MobileV2 Convolution Neuron Network (CNN) model and the author adjusted the top layers to match the FER classification task, the special inverted residual blocks in the MobileV2 accelerate the training process while ensuring the high accuracy. Since the data were all labeled, this study applied model fine-tuning and froze the weights of the first few layers in the model which were trained to detect the special features in the images. Thus, by adjusting the weights of the fully connected layers, the model successfully transferred to a similar domain. Experimental results indicated that after applying the model fine-tuning, the FER model performed much better while recognizing colorful images of faces from different human races and the new model reduced the bias created by the previous training dataset.

Copyright

© 2024 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Zhisong Liu
PY  - 2024
DA  - 2024/10/16
TI  - Fine-tuning Technologies for Reducing the FER Bias Across Various Distributions
BT  - Proceedings of the 2024 2nd International Conference on Image, Algorithms and Artificial Intelligence (ICIAAI 2024)
PB  - Atlantis Press
SP  - 921
EP  - 929
SN  - 2352-538X
UR  - https://doi.org/10.2991/978-94-6463-540-9_92
DO  - 10.2991/978-94-6463-540-9_92
ID  - Liu2024
ER  -