S2-Net: Self-Supervision Guided Feature Representation Learning for Cross-Modality Images

Dear Editor, This letter focuses on combining the respective advantages of cross-modality images which can compensate for lack information in single modality. Meanwhile, due to great appearance differences between image pairs, it often fails make feature representations correspondences as close possible. In this letter, we design a representation learning network, S2-Net, is based recently successful detect-and-describe pipeline, originally proposed visible but adapted work with pairs. Extensive experiments show that our elegant formulation combined optimization supervised and self-supervised outperforms state-of-the-arts three cross-modal datasets.