S10: Attentive Models and Visual Attention in Computer Vision and AI

Lorenzo Baraldi
Marcella Cornia

Università degli Studi di Modena e Reggio Emilia, Italia

Abstract

Computer vision has always been a rapidly evolving field with various real-world applications. In the annals of its history, a withstanding problem has been modeling and exploiting the visual attention mechanism. From there, the computational models of visual attention have contributed to formation of various subproblems in the field, including saliency modeling and eye-fixation prediction, which also have gained a fundamental role in the standard computer vision pipeline. Nowadays, with the surge of attentive and Transformer-based models, the modelling of attention has grown significantly and is the pillar of cutting-edge research in vision and artificial intelligence. In this context, this research area also contributes with new architectures which are candidates to replace the convolutional operator, as testified by recent works which perform image classification using attention-based architectures.

The special session on “Attentive Models and Visual Attention in Computer Vision and AI” fosters the submission of innovative research works which relate to the study of human attention and to the usage of attention mechanisms in the development of deep learning architectures. Works employing traditional attentive operations or employing novel Transformer-based architectures are encouraged. Submissions are solicited from all areas of AI, including vision, natural language and multi-modal integration. Works tackling the computational issues behind the usage of such models, and which tackle the development of optimization solutions, are also encouraged. The special session also favours a positive criticism in the role of attention with respect to more traditional approaches. Quantitative comparisons of existing solutions and datasets are also welcome to raise awareness on the topic.

The topics of interest include but are not limited to

• Saliency prediction and saliency detection
• Visualization of attentive maps for Explainability of Deep Networks
• Applications of attentive operators in the design of Convolutional Neural Networks
• Applications of attentive operators for Natural Language Processing and Understanding
• Usage of Transformer-based or attention-based models in Vision
• Integration of multiple modalities through attentive models
• Computational issues in attention-based models