On robust cross-view consistency in self-supervised monocular depth estimation
Researchers propose new cross-view consistency losses to enhance self-supervision signal, achieving superior results in monocular depth estimation. The method leverages temporal coherence in depth feature space and 3D voxel space to mitigate the side effects of challenging cases.
SourceBeijing Zhongke Journal Publising Co. Ltd.·JournalMachine Intelligence Research·DateJun 17, 2024