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 Plus annual

Presents the access of premium content as premium feature

Premium Content

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 annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Depth estimation is a fundamental issue in 4-D light field processing andanalysis. Although recent supervised learning-based light field depthestimation methods have significantly improved the accuracy and efficiency oftraditional optimization-based ones, these methods rely on the training overlight field data with ground-truth depth maps which are challenging to obtainor even unavailable for real-world light field data. Besides, due to theinevitable gap (or domain difference) between real-world and synthetic data,they may suffer from serious performance degradation when generalizing themodels trained with synthetic data to real-world data. By contrast, we proposean unsupervised learning-based method, which does not require ground-truthdepth as supervision during training. Specifically, based on the basicknowledge of the unique geometry structure of light field data, we present anocclusion-aware strategy to improve the accuracy on occlusion areas, in whichwe explore the angular coherence among subsets of the light field views toestimate initial depth maps, and utilize a constrained unsupervised loss tolearn their corresponding reliability for final depth prediction. Additionally,we adopt a multi-scale network with a weighted smoothness loss to handle thetextureless areas. Experimental results on synthetic data show that our methodcan significantly shrink the performance gap between the previous unsupervisedmethod and supervised ones, and produce depth maps with comparable accuracy totraditional methods with obviously reduced computational cost. Moreover,experiments on real-world datasets show that our method can avoid the domainshift problem presented in supervised methods, demonstrating the greatpotential of our method.

Occlusion-Aware Unsupervised Learning of Depth From 4-D Light Fields