Self-Supervised deep homography estimation with invertibility constraints
Pattern Recognition Letters(2019)
摘要
•We propose a self-supervised regression network for homography estimation by generating synthetic images in a cyclic way.•We propose the invertibility loss which contributes to avoid over-fitting and improve the homography estimation accuracy.•Experiments on MS-COCO dataset show that our SSR-Net outperforms state-of-the-art traditional and self-supervised methods.
更多查看译文
关键词
Homography estimation,Self-Supervised deep learning,Invertibility constraint,Spatial pyramid pooling
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络