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

Monocular depth estimation is challenging due to its inherent ambiguity andill-posed nature, yet it is quite important to many applications. While recentworks achieve limited accuracy by designing increasingly complicated networksto extract features with limited spatial geometric cues from a single RGBimage, we intend to introduce spatial cues by training a teacher network thatleverages left-right image pairs as inputs and transferring the learned 3Dgeometry-aware knowledge to the monocular student network. Specifically, wepresent a novel knowledge distillation framework, named ADU-Depth, with thegoal of leveraging the well-trained teacher network to guide the learning ofthe student network, thus boosting the precise depth estimation with the helpof extra spatial scene information. To enable domain adaptation and ensureeffective and smooth knowledge transfer from teacher to student, we apply bothattention-adapted feature distillation and focal-depth-adapted responsedistillation in the training stage. In addition, we explicitly model theuncertainty of depth estimation to guide distillation in both feature space andresult space to better produce 3D-aware knowledge from monocular observationsand thus enhance the learning for hard-to-predict image regions. Our extensiveexperiments on the real depth estimation datasets KITTI and DrivingStereodemonstrate the effectiveness of the proposed method, which ranked 1st on thechallenging KITTI online benchmark.

ADU-Depth: Attention-based Distillation with Uncertainty Modeling for Depth Estimation