Research map / Detection and segmentation / 3D detection and driving
MemDistill: Distilling LiDAR Knowledge into Memory for Camera-Only 3D Object Detection
Donghyeon Kwon, Youngseok Yoon, Hyeongseok Son, Suha Kwak
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add student fine-tuning after memory distillation (ablation) | MemDistill · nuScenes val (R50) | mAP: 0.348 -> 0.425 | Improved |
| add student fine-tuning after memory distillation (ablation) | MemDistill · nuScenes val (R50) | NDS: 0.475 -> 0.531 | Improved |
| add memory distillation before fine-tuning (ablation) | MemDistill · nuScenes val (R50) | mAP: 0.382 -> 0.425 | Improved |
| add memory distillation before fine-tuning (ablation) | MemDistill · nuScenes val (R50) | NDS: 0.498 -> 0.531 | Improved |
| use two-stage training instead of end-to-end (ablation) | MemDistill · nuScenes val (R50) | mAP: 0.369 -> 0.425 | Improved |
| use two-stage training instead of end-to-end (ablation) | MemDistill · nuScenes val (R50) | NDS: 0.475 -> 0.531 | Improved |
| add scene-dependent memory retriever instead of scene-agnostic encoder (ablation) | MemDistill · nuScenes val (R50) | mAP: 0.392 -> 0.425 | Improved |
| add scene-dependent memory retriever instead of scene-agnostic encoder (ablation) | MemDistill · nuScenes val (R50) | NDS: 0.499 -> 0.531 | Improved |
| use LiDAR teacher neck and head in student instead of camera neck and head (ablation) | MemDistill student · nuScenes val feature discrepancy | L2 distance: 501.1 -> 395.7 | Improved |
| add memory distillation to LiDAR neck/head student (ablation) | MemDistill student · nuScenes val feature discrepancy | L2 distance: 395.7 -> 344.3 | Improved |
| add fine-tuning on top of memory distillation (ablation) | MemDistill student · nuScenes val feature discrepancy | L2 distance: 344.3 -> 333.9 | Improved |
| add memory retrieval and fusion (ablation) | camera BEV features · nuScenes val feature discrepancy | L2 distance: 975.2 -> 604.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- LabelDistill — LabelDistill: Label-guided Cross-modal Knowledge Distillation for Camera-based 3D Object Detection (ECCV 2024) · better on 4 of 4 shared results
- PolarFormer — PolarFormer: Multi-Camera 3D Object Detection with Polar Transformer (AAAI 2023) · better on 2 of 2 shared results
- BEVFormer v2 — BEVFormer v2: Adapting Modern Image Backbones to Bird's-Eye-View Recognition via Perspective Supervision (CVPR 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Advancing 3D Object Detection With Depth-Aware Spatial Knowledge Distillation (TPAMI 2025 · similarity 0.90)
- DistillBEV: Boosting Multi-Camera 3D Object Detection with Cross-Modal Knowledge Distillation (ICCV 2023 · similarity 0.90)
- Once Detected, Never Lost: Surpassing Human Performance in Offline LiDAR based 3D Object Detection (ICCV 2023 · similarity 0.90)
- Towards Universal LiDAR-Based 3D Object Detection by Multi-Domain Knowledge Transfer (ICCV 2023 · similarity 0.90)
- MAD: Memory-Augmented Detection of 3D Objects (CVPR 2025 · similarity 0.89)
- PromptDet: A Lightweight 3D Object Detection Framework with LiDAR Prompts (AAAI 2025 · similarity 0.88)
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.
