phase congruency-based loss

UMHE: Unsupervised Multispectral Homography Estimation

In this paper, we present a comprehensive study on unsupervised multispectral homography estimation. We introduce a FLIR-corresponding dataset that can be used to evaluate image registration methods. Our model running at 151 FPS achieved state-of-the-art performance on the proposed FLIR-correspondence dataset and showed good generalization without retraining on the M3FD dataset.