Self-supervised monocular depth learning from unknown cameras: Leveraging the power of raw data

Download PDF
Open Webpage

Xiaofei Qin, Yongchao Zhu, Lin Wang, Xuedian Zhang, Changxiang He, Qiulei Dong

Published: 2025, Last Modified: 05 Nov 2025Image Vis. Comput. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Joint residual pose-reprojection optimization handles noisy extreme rotations.•NIM enhances feature via local feature fusion with low-cost block computation.•Distilled lightweight encoder cuts params while retaining scene understanding.•cross-video training improves unseen env accuracy.