Sign in to view source links and access this dataset
Description
105,268 PNG images of face illusion stimuli created for 26,317 identities, forming a large-scale benchmark for testing human-vs-deep neural network alignment on configural face processing. The dataset, created by Enhui-1, is constructed entirely with classical computer vision techniques like dlib and OpenCV, with no generative AI used. It was last updated on Hugging Face in May 2026.
Use Cases
Testing human-vs-DNN alignment on configural face processing based on the holistic-face illusion stimuli.
Evaluating paired EEG decoders based on the benchmark's structured stimuli.
Researching holistic face perception using stimuli created with pure classical computer vision methods.
Benchmarking computer vision models on face processing tasks without generative AI artifacts.
Strengths
Large scale with 105,268 images derived from 26,317 unique identities.
Stimuli created with pure classical computer vision (dlib, OpenCV), avoiding potential biases from generative AI.
Specifically designed for paired EEG-decoder evaluation, providing a targeted resource.
Limitations
Column-level documentation is absent; field semantics must be inferred after download.
Description metadata is limited; actual data quality and structure require manual inspection after download.