Connectivity-inspired network for context-aware recognition

The aim of this paper is threefold. We inform the AI practitioner about thehuman visual system with an extensive literature review; we propose a novelbiologically motivated neural network for image classification; and, finally,we present a new plug-and-play module to model context awareness. We focus onthe effect of incorporating circuit motifs found in biological brains toaddress visual recognition. Our convolutional architecture is inspired by theconnectivity of human cortical and subcortical streams, and we implementbottom-up and top-down modulations that mimic the extensive afferent andefferent connections between visual and cognitive areas. Our ContextualAttention Block is simple and effective and can be integrated with anyfeed-forward neural network. It infers weights that multiply the feature mapsaccording to their causal influence on the scene, modeling the co-occurrence ofdifferent objects in the image. We place our module at different bottlenecks toinfuse a hierarchical context awareness into the model. We validated ourproposals through image classification experiments on benchmark data and founda consistent improvement in performance and the robustness of the producedexplanations via class activation. Our code is available at https://github.com/gianlucarloni/CoCoReco.