Connectivity-Inspired Network for Context-Aware Recognition

Abstract

In this paper, we focus on the effect of incorporating circuit motifs found in biological brains to address visual recognition. After an extensive literature review of the human visual system, we propose CoCoReco , a novel co nnectivity-inspired and co ntext-aware reco gnition network. Motivated by the connectivity of human (sub)cortical streams, we implement bottom-up and top-down modulations that mimic the extensive connections between visual and cognitive areas. Moreover, we present the Contextual Attention Block, a new plug-and-play module that can be integrated with any feed-forward neural network to model visual context. It infers weights that multiply the feature maps according to their causal influence on the scene, modeling the co-occurrence of objects. We place our module at different bottlenecks to infuse a hierarchical context awareness. We validate CoCoReco on image classification experiments on benchmark data and find consistent improvements in performance and robustness of the produced explanations. Find our code at https://github.com/gianlucarloni/CoCoReco .