Research map / Detection and segmentation / 3D detection and driving
Domain Generalization of 3D Object Detection by Density-Resampling
Shuangzhi Li, Lei Ma, Xingyu Li
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 PDDA density-resampling augmentation during source training (ablation) | VoxelRCNN source-only · NuScenes -> KITTI | mAP BEV/3D: 35.34/18.5 -> 41.07/24.08 | Improved |
| add auxiliary self-supervised scene restoration (multi-task learning) (ablation) | VoxelRCNN + PDDA · NuScenes -> KITTI | mAP BEV/3D: 41.07/24.08 -> 41.35/24.12 | No clear effect |
| add test-time adaptation using the restoration task (ablation) | VoxelRCNN + PDDA + MTL · NuScenes -> KITTI | mAP BEV/3D: 41.35/24.12 -> 42.17/25.15 | Improved |
| add test-time adaptation using the restoration task (ablation) | VoxelRCNN + PDDA + MTL · Waymo -> NuScenes | mAP BEV/3D: 17.09/11.17 -> 17.22/11.26 | Improved |
| replace density-downsampling test-time training with KNN-masking TTT (ablation) | test-time adaptation · NuScenes -> KITTI | FPS: 4.23 -> 0.52 | Hurt |
| add perceptual loss to MSE restoration loss (ablation) | test-time adaptation · NuScenes -> KITTI | mAP BEV/3D and FPS: 42.17/25.15 -> 42.21/25.16; FPS 4.23 -> 0.89 | No clear effect |
| increase test-time update iterations beyond optimum (ablation) | test-time adaptation · NuScenes -> KITTI and Waymo -> NuScenes | mAP | Hurt |
| reduce test-time iterations to N_iter=1 for real-time speed (ablation) | test-time adaptation · NuScenes -> KITTI | mAP BEV/3D: 42.17/25.15 -> 41.41/24.19 | No clear effect |
| apply Ours versus source-only on domains without density gap | Ours · Waymo -> KITTI | mAP BEV/3D: 65.41/46.79 -> 65.47/46.67 | No clear effect |
| use PDDA instead of UDA augmentation RBRS that needs target info (ablation) | source training augmentation · Waymo -> NuScenes | mAP BEV/3D: 17.33/11.20 -> 16.83/11.04 | Mixed |
Closest papers by meaning
- Unified Domain Generalization and Adaptation for Multi-View 3D Object Detection (NeurIPS 2024 · similarity 0.92)
- Improving Single Domain-Generalized Object Detection: A Focus on Diversification and Alignment (CVPR 2024 · similarity 0.90)
- Object-Aware Domain Generalization for Object Detection (AAAI 2024 · similarity 0.90)
- Unbiased Faster R-CNN for Single-source Domain Generalized Object Detection (CVPR 2024 · similarity 0.89)
- Towards Single-Source Domain Generalized Object Detection via Causal Visual Prompts (NeurIPS 2025 · similarity 0.89)
- One for All: Multi-Domain Joint Training for Point Cloud Based 3D Object Detection (NeurIPS 2024 · similarity 0.89)
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.
