AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

Representation Disparity-aware Distillation for 3D Object Detection

Yanjing Li, Sheng Xu, Mingbao Lin, Jihao Yin, Baochang Zhang, Xianbin Cao

ICCV 2023 · Label-efficient and robust learning › Knowledge distillation · approach: Knowledge distillation for detectors · 10 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace Hint (full neck feature distillation) with RDD region selection (ablation)CP-Voxel-S · nuScenesmAP: 52.9 -> 57.1Improved
use Hint region selection instead of no distillation (ablation)CP-Voxel-S · nuScenesmAP: 54.0 -> 52.9Hurt
add feature-level RDD loss alone (ablation)CP-Voxel-S · nuScenesmAP: 54.0 -> 56.8Improved
add logit-level RDD loss alone (ablation)CP-Voxel-S · nuScenesmAP: 54.0 -> 57.0Improved
combine feature-level and logit-level RDD losses (ablation)CP-Voxel-S · nuScenesmAP: 57.0 -> 57.1Improved
set lambda to 0 so all region pairs are distilled equally (ablation)RDD · nuScenes CP-Voxel-SmAPHurt
enlarge distillation loss weights alpha1 and alpha2 beyond 0.2 (ablation)RDD · nuScenes CP-Voxel-SmAPHurt
distill with RDD versus previous 3D KD (PP Logit KD)CP-Voxel-XXS · nuScenesmAP: 47.9 -> 49.4Improved
distill with RDD versus PP Logit KD on the smallest nuScenes voxel studentCP-Voxel-XS · nuScenesmAP: 53.5 -> 54.0Improved
distill CP-Voxel-S with RDD, student surpasses its teacherCP-Voxel-S · nuScenesmAP: 56.6 -> 57.1Improved
apply 2D feature distillation baselines (GID, LAD) to 3D detectorsCP-Voxel-S · nuScenesmAP: 54.0 -> 54.1 (GID-L), 53.0 (LAD)No clear effect
distill with RDD versus PointDistiller on KITTIPointPillars-S · KITTI val moderatemAP@R40: 62.3 -> 63.0Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy