Scholarlink

This study explores the relationship between Deep Neural Networks (DNNs) and human visual processing through Skye’s Oblique Grating Illusion. By analyzing perceptual angle data from human subjects and DNN responses to the illusion, complemented with fMRI-based Region of Interest (ROI) analysis, we uncover both distinct differences and similarities. Our findings demonstrate that while DNNs initially show human-like responses to visual illusions at early processing stages, they diverge in deeper processing layers. Additionally, the significant activation of the V2 region in human subjects highlights its critical role in illusion processing. Our research further reveals that models like Vgg19, which closely correlate with human perception, exhibit unique feature activation patterns, emphasizing the complexity of emulating human visual processing. Future research should expand on these insights to further elucidate the intricate relationship between DNNs and human visual processing.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Perceptual Parallels and Divergences: Analyzing Slant Illusion Response in Deep Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Hongtao Zhang,
  • Shinichi Yoshida

摘要

This study explores the relationship between Deep Neural Networks (DNNs) and human visual processing through Skye’s Oblique Grating Illusion. By analyzing perceptual angle data from human subjects and DNN responses to the illusion, complemented with fMRI-based Region of Interest (ROI) analysis, we uncover both distinct differences and similarities. Our findings demonstrate that while DNNs initially show human-like responses to visual illusions at early processing stages, they diverge in deeper processing layers. Additionally, the significant activation of the V2 region in human subjects highlights its critical role in illusion processing. Our research further reveals that models like Vgg19, which closely correlate with human perception, exhibit unique feature activation patterns, emphasizing the complexity of emulating human visual processing. Future research should expand on these insights to further elucidate the intricate relationship between DNNs and human visual processing.