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

To achieve accurate and low-cost 3D object detection, existing methodspropose to benefit camera-based multi-view detectors with spatial cues providedby the LiDAR modality, e.g., dense depth supervision and bird-eye-view (BEV)feature distillation. However, they directly conduct point-to-point mimickingfrom LiDAR to camera, which neglects the inner-geometry of foreground targetsand suffers from the modal gap between 2D-3D features. In this paper, wepropose the learning scheme of Target Inner-Geometry from the LiDAR modalityinto camera-based BEV detectors for both dense depth and BEV features, termedas TiG-BEV. First, we introduce an inner-depth supervision module to learn thelow-level relative depth relations between different foreground pixels. Thisenables the camera-based detector to better understand the object-wise spatialstructures. Second, we design an inner-feature BEV distillation module toimitate the high-level semantics of different keypoints within foregroundtargets. To further alleviate the BEV feature gap between two modalities, weadopt both inter-channel and inter-keypoint distillation for feature-similaritymodeling. With our target inner-geometry distillation, TiG-BEV can effectivelyboost BEVDepth by +2.3% NDS and +2.4% mAP, along with BEVDet by +9.1% NDS and+10.3% mAP on nuScenes val set. Code will be available athttps://github.com/ADLab3Ds/TiG-BEV.

TiG-BEV: Multi-view BEV 3D Object Detection via Target Inner-Geometry Learning