SBDet: A Symmetry-Breaking Object Detector via Relaxed Rotation-Equivariance

• DOI: 10.48550/arxiv.2408.11760
• Publication Date: 2024-08-21
• Authors: Zhiqiang Wu; Yingjie Liu; Hanlin Dong; Xuan Tang; Jian Yang; Bo Jin; Mingsong Chen; Xian Wei
• Keywords: FOS: Computer and information sciences; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition

Introducing Group Equivariant Convolution (GConv) empowers models to explore symmetries hidden in visual data, improving their performance. However, real-world scenarios, objects or scenes often exhibit perturbations of a symmetric system, specifically deviation from architecture, which can be characterized by non-trivial action symmetry group, known as Symmetry-Breaking. Traditional GConv methods are limited the strict operation rules group space, only ensuring features remain strictly equivariant under transformations, making it difficult adapt Symmetry-Breaking non-rigid transformations. Motivated this, we introduce novel Relaxed Rotation (R2GConv) with our defined Rotation-Equivariant $\mathbf{R}_4$. Furthermore, propose Network (R2Net) backbone and further develop Object Detector (SBDet) for 2D object detection built upon it. Experiments demonstrate effectiveness proposed R2GConv natural image classification tasks, SBDet achieves excellent performance tasks improved generalization capabilities robustness.