Evaluating XR Beyond WEIRD Population: Study Replication for Video-Based Self-Avatars

Full body video-based self-avatars allow users to see themselves while immersed in extended reality applications. A recent study suggests that deep learning-based segmentation solutions can increase the embodiment and visual quality of certain body parts, such as legs, in comparison with virtual avatars or chromatic segmentation video-based solutions. However, further research is required to be more certain about these statements, since like most related works, this previous study was conducted with university students (54). In this work, we replicate this study with a local population: a subset of 33 participants from different backgrounds recruited by a specialized company. When comparing results between both populations, we observe: $i$) the local population tends to observe statistically significance between conditions for more factors than the university population, ii) there are also differences of significance when analyzing embodiment and visual quality per age or gender iii) there might be additional dimension factors such as color skin that influence the algorithm results, or the educational background/job profile affecting people's responses. We finish the paper with some reflection on current limitations of the study and lessons learnt on how to properly face a research study with a broader population. While this is a good starting point, further efforts are necessary to fully address WEIRD biases in research populations.