Enhui-1/HoloFaceIllusion-Bench-EEG
收藏资源简介:
HoloFaceIllusion-Bench-EEG是一个大规模基准数据集,专注于整体面部错觉刺激,用于测试人类与深度神经网络(DNN)在配置面部处理方面的对齐性,并支持配对的脑电图(EEG)解码器评估。数据集包含三个经典视觉错觉范式:Thatcher错觉(基于Thompson, 1980)、Part-Whole错觉(基于Tanaka & Farah, 1993)和Composite错觉(基于Young et al., 1987)。所有图像均使用传统计算机视觉技术生成,包括dlib地标检测、MediaPipe Face Mesh、InsightFace性别/年龄识别、OpenCV泊松克隆和Reinhard LAB颜色转移,未使用任何神经网络或生成式AI。数据集总计包含1,107,972张图像,基于FFHQ数据集派生,遵循CC BY-NC-SA 4.0许可,适用于学术和非商业用途。具体来说,Thatcher部分包含26,317个身份,每个身份有4个条件(正常直立、Thatcher化直立、正常倒置、Thatcher化倒置),共105,268张图像;Part-Whole部分包含188,988个案例(模板×目标×特征),每个案例4个条件,共755,952张图像;Composite部分包含61,688个案例(模板×捐赠者),每个案例4个条件,共246,752张图像。数据集旨在为心理物理学、计算机视觉和神经科学研究提供标准化刺激材料。
HoloFaceIllusion-Bench-EEG is a large-scale benchmark of holistic-face illusion stimuli for testing human-vs-DNN alignment on configural face processing and for paired EEG-decoder evaluation. It includes three classical visual illusion paradigms: Thatcher illusion (based on Thompson, 1980), Part-Whole illusion (based on Tanaka & Farah, 1993), and Composite illusion (based on Young et al., 1987). All images are generated entirely using classical computer vision techniques, including dlib landmarks, MediaPipe Face Mesh, InsightFace gender/age detection, OpenCV Poisson cloning, and Reinhard LAB color transfer, with no neural networks or generative AI used. The dataset totals 1,107,972 images, derived from the FFHQ dataset, and is licensed under CC BY-NC-SA 4.0 for academic and non-commercial use. Specifically, the Thatcher section comprises 26,317 identities with 4 conditions each (upright normal, upright thatchered, inverted normal, inverted thatchered), resulting in 105,268 images; the Part-Whole section includes 188,988 cases (template × target × feature) with 4 conditions each, totaling 755,952 images; and the Composite section contains 61,688 cases (template × donor) with 4 conditions each, amounting to 246,752 images. The dataset serves as a standardized stimulus resource for psychophysics, computer vision, and neuroscience research.




