Ying Hu^1,2 (), Ruyu Pan^1,2, Yaqi Xiao^1,2, Zihan Zhu^1,2, Geraldine Jeckeln³, Xiaolan Fu⁴; ¹State Key Laboratory of Brain and Cognitive Science, Institute of Psychology, Chinese Academy of Sciences, Beijing, 100101, China, ²Department of Psychology, University of Chinese Academy of Sciences, Beijing, 100049, China, ³The University of Texas at Dallas, School of Behavioral and Brain Sciences, ⁴School of Psychology, Shanghai Jiaotong University, Shanghai 200030, China

Face perception research typically focuses on identity differences, primarily using White subjects. This approach often neglects how appearances can change across images and the diversity of other racial groups. We aim to develop a dataset to address these gaps and to assess the effectiveness of the dataset in facilitating the exploration of within-model variability under a range of conditions. The Chinese Face and Body Dataset (CFBD) offers a publicly accessible collection of 2,195 images from 117 Asian models. This dataset includes both standardized lab photos and personal photos taken in natural settings at various times. Each image was manually annotated for eight objective attributes: facial expressions, face views, body views, postures, photo types, environments, image quality, and accessories. Participants rated cropped face images from the CFBD for two social traits (attractiveness, trustworthiness) and one identity-relevant trait (distinctiveness). Attractiveness and distinctiveness showed higher within-model variability than between-model variability, as indicated by Wilcoxon rank-sum tests (W = 5733, p = .022 for attractiveness; W = 7137, p < .001 for distinctiveness), whereas trustworthiness did not (W = 11700, p = .547). Personal photos received higher ratings than lab photos for attractiveness (t(88) = 9.676, p < .001), distinctiveness (t(88) = 9.618, p < .001), and trustworthiness (t(88) = 6.993, p < .001), underscoring the impact of type of photos on face perceptions. These findings suggest that perceived attractiveness and distinctiveness are more dependent on cues that vary across images, whereas trustworthiness is relatively dependent on cues that remain consistent. By enabling the exploration of within-model variability under diverse conditions, the CFBD significantly enhances face perception research, particularly for a previously underrepresented group in face perception.

Acknowledgements: National Natural Science Foundation of China (NSFC) grant No. 32200866 (YH). National Natural Science Foundation of China (NSFC) grant No. 62276259 (YH).