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

This paper investigates the geometric consistency for monocular 3D objectdetection, which suffers from the ill-posed depth estimation. We first conducta thorough analysis to reveal how existing methods fail to consistentlylocalize objects when different geometric shifts occur. In particular, wedesign a series of geometric manipulations to diagnose existing detectors andthen illustrate their vulnerability to consistently associate the depth withobject apparent sizes and positions. To alleviate this issue, we propose fourgeometry-aware data augmentation approaches to enhance the geometricconsistency of the detectors. We first modify some commonly used dataaugmentation methods for 2D images so that they can maintain geometricconsistency in 3D spaces. We demonstrate such modifications are important. Inaddition, we propose a 3D-specific image perturbation method that employs thecamera movement. During the augmentation process, the camera system with thecorresponding image is manipulated, while the geometric visual cues for depthrecovery are preserved. We show that by using the geometric consistencyconstraints, the proposed augmentation techniques lead to improvements on theKITTI and nuScenes monocular 3D detection benchmarks with state-of-the-artresults. In addition, we demonstrate that the augmentation methods are wellsuited for semi-supervised training and cross-dataset generalization.

Exploring Geometric Consistency for Monocular 3D Object Detection