Z. Wang, D. Liu, S. Chang, F. Dolcos, D. Beck, and T. Huang,

"Image Aesthetics Assessment using Deep Chatterjee's Machine'',

In Proceedings of  International Joint Conference on Neural Networks (IJCNN), 2017. [Earlier arXiv verion]

Scientific Basis: Chatterjee‘s visual neuroaesthetics model

  • The human brain works as a multi-leveled system.
  • For the visual sensory input, a variety of relevant feature dimensions (most relevant features) are first targeted.
  • A set of parallel pathways abstract the visual input. Each pathway processes the input into an attribute on a specific feature dimension.
  • The high-level association module transforms all attributes into an aesthetics decision.

A “biologically-interpretable”, Brain-inspired Deep Network (BDN)

Examples of BDN classification results (AVA): (a) high-quality; (b) low-quality .

Zhangyang (Atlas) Wang​