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

Recent advances in 3D object detection (3DOD) have obtained remarkably strongresults for LiDAR-based models. In contrast, surround-view 3DOD models based onmultiple camera images underperform due to the necessary view transformation offeatures from perspective view (PV) to a 3D world representation which isambiguous due to missing depth information. This paper introduces X$^3$KD, acomprehensive knowledge distillation framework across different modalities,tasks, and stages for multi-camera 3DOD. Specifically, we propose cross-taskdistillation from an instance segmentation teacher (X-IS) in the PV featureextraction stage providing supervision without ambiguous error backpropagationthrough the view transformation. After the transformation, we apply cross-modalfeature distillation (X-FD) and adversarial training (X-AT) to improve the 3Dworld representation of multi-camera features through the information containedin a LiDAR-based 3DOD teacher. Finally, we also employ this teacher forcross-modal output distillation (X-OD), providing dense supervision at theprediction stage. We perform extensive ablations of knowledge distillation atdifferent stages of multi-camera 3DOD. Our final X$^3$KD model outperformsprevious state-of-the-art approaches on the nuScenes and Waymo datasets andgeneralizes to RADAR-based 3DOD. Qualitative results video athttps://youtu.be/1do9DPFmr38.

X$^3$KD: Knowledge Distillation Across Modalities, Tasks and Stages for Multi-Camera 3D Object Detection