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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The best way to combine the results of deep learning with standard 3Dreconstruction pipelines remains an open problem. While systems that pass theoutput of traditional multi-view stereo approaches to a network forregularisation or refinement currently seem to get the best results, it may bepreferable to treat deep neural networks as separate components whose resultscan be probabilistically fused into geometry-based systems. Unfortunately, theerror models required to do this type of fusion are not well understood, withmany different approaches being put forward. Recently, a few systems haveachieved good results by having their networks predict probabilitydistributions rather than single values. We propose using this approach to fusea learned single-view depth prior into a standard 3D reconstruction system.  Our system is capable of incrementally producing dense depth maps for a setof keyframes. We train a deep neural network to predict discrete, nonparametricprobability distributions for the depth of each pixel from a single image. Wethen fuse this "probability volume" with another probability volume based onthe photometric consistency between subsequent frames and the keyframe image.We argue that combining the probability volumes from these two sources willresult in a volume that is better conditioned. To extract depth maps from thevolume, we minimise a cost function that includes a regularisation term basedon network predicted surface normals and occlusion boundaries. Through a seriesof experiments, we demonstrate that each of these components improves theoverall performance of the system.

Towards the Probabilistic Fusion of Learned Priors into Standard Pipelines for 3D Reconstruction