Channel attention deep learning for renaissance art analysis

The integration of deep learning and computer vision technologies has revolutionized the study of paintings, offering new insights into historical and cultural contexts. However, accurately identifying objects in paintings remains a challenge due to the inherent complexities of various artistic styles. This research focuses on the Renaissance period, specifically examining the Early Renaissance, High Renaissance, Tenebrism, and Academicism styles, to compare human portrayal through the object detection of faces and hands. Our contributions include a detailed analysis of visual characteristics across these styles and the development of a channel-wise attention mechanism within deep learning object detection. Our findings reveal that integrating the Squeeze-and-Excitation (SE) block with a multi-scale model like RetinaNet significantly enhances the model’s robustness by addressing issues related to low color palette and color imbalance. The results demonstrate marked improvements in detection accuracy and adaptability within the complex context of Renaissance art, underscoring the effectiveness of the SE-enhanced RetinaNet model in advancing object detection in stylized images.