University of Twente Student Theses

Login

Investigating Cross-cultural Generalizability of Facial Emotion Recognition with Multi-dataset Training

Chen, Yulin (2024) Investigating Cross-cultural Generalizability of Facial Emotion Recognition with Multi-dataset Training.

[img] PDF
2MB
Abstract:Facial emotion recognition (FER) is among computer vision’s most complex fields and has practical uses in human-computer interaction (HCI) and psychology. Currently, FER models are trained on datasets dominated by a singular ethnicity, for instance, AffectNet [19]. AffectNet is a popular dataset for FER tasks with over 400,000 manually annotated samples, and has 64.4% of its training data represented by White subjects. Consequently, the accuracy is often limited when the model is deployed in the real world, where the population is much more culturally diverse. This research will investigate the impact of augmenting FER datasets with EiLA (Emotions in LatAm Dataset), a newly curated emotion recognition in-the-wild dataset consisting of video recordings of Latin Americans and their facial expressions, on the accuracy and performance of well-known FER models. The study begins with dataset preparation, where different-sized portions of the EiLA dataset are integrated with two existing FER datasets. Next, the integrated datasets are used to train the chosen neural network, followed by testing and metrics evaluation using ground truth labels from the EiLA dataset. Finally, the analysis of results will be interpreted to determine whether augmenting the cultural diversity of datasets positively impacts the efficacy of FER models.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:https://purl.utwente.nl/essays/100771
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page