English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Remarkable progress has been made in self-supervised monocular depthestimation (SS-MDE) by exploring cross-view consistency, e.g., photometricconsistency and 3D point cloud consistency. However, they are very vulnerableto illumination variance, occlusions, texture-less regions, as well as movingobjects, making them not robust enough to deal with various scenes. To addressthis challenge, we study two kinds of robust cross-view consistency in thispaper. Firstly, the spatial offset field between adjacent frames is obtained byreconstructing the reference frame from its neighbors via deformable alignment,which is used to align the temporal depth features via a Depth FeatureAlignment (DFA) loss. Secondly, the 3D point clouds of each reference frame andits nearby frames are calculated and transformed into voxel space, where thepoint density in each voxel is calculated and aligned via a Voxel DensityAlignment (VDA) loss. In this way, we exploit the temporal coherence in bothdepth feature space and 3D voxel space for SS-MDE, shifting the"point-to-point" alignment paradigm to the "region-to-region" one. Comparedwith the photometric consistency loss as well as the rigid point cloudalignment loss, the proposed DFA and VDA losses are more robust owing to thestrong representation power of deep features as well as the high tolerance ofvoxel density to the aforementioned challenges. Experimental results on severaloutdoor benchmarks show that our method outperforms current state-of-the-arttechniques. Extensive ablation study and analysis validate the effectiveness ofthe proposed losses, especially in challenging scenes. The code and models areavailable at https://github.com/sunnyHelen/RCVC-depth.

On Robust Cross-View Consistency in Self-Supervised Monocular Depth Estimation