Deformable Feature Alignment and Refinement for moving infrared small target detection

Dengyan Luo, Yanping Xiang, Hu Wang, Luping Ji, Shuai Li, Mao Ye

Abstract: Highlights•We are the first to use deformable convolution for moving infrared small target detection, and extensive experimental results demonstrate the effectiveness of our proposed Deformable Feature Alignment and Refinement (DFAR) method.•A temporal deformable alignment module is proposed by explicitly aligning adjacent frames with the current frame at the feature level to mine temporal information.•A feature refinement module is designed to adaptively fuse and refine useful temporal information from the aligned features.•A motion compensation loss is introduced to improve the temporal alignment.