Research map / Detection and segmentation / 3D detection and driving
Weakly Supervised 3D Object Detection via Multi-Level Visual Guidance
Kuan-Chih Huang, Yi-Hsuan Tsai, Ming-Hsuan Yang
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 output-level 2D-3D box guidance (GIoU between projected 3D box and 2D box) (ablation) | VG-W3D baseline · KITTI val Car AP3D Mod. | AP3D: 74.00 -> 76.42 | Improved |
| Add training-level guidance (score-consistent pseudo-label refinement) (ablation) | VG-W3D baseline · KITTI val Car AP3D Mod. | AP3D: 74.00 -> 77.75 | Improved |
| Add feature-level objectness guidance alone (ablation) | VG-W3D baseline · KITTI val Car AP3D Mod. | AP3D: 74.00 -> 74.29 | Improved |
| Combine feature, output and training guidance (ablation) | VG-W3D · KITTI val Car AP3D Mod. | AP3D: 74.00 -> 78.89 | Improved |
| Use KL objectness loss with image branch instead of omitting the image-domain objectness (ablation) | Feature-level guidance · KITTI val Car AP3D Mod. | AP3D: 74.61 -> 76.71 | Improved |
| Use KL objectness distillation instead of L2 feature mimicking (ablation) | Feature-level guidance · KITTI val Car AP3D Mod. | AP3D: 75.69 -> 76.71 | Improved |
| Use self-supervised segmentation mask instead of 2D box mask for foreground supervision (ablation) | Feature-level guidance · KITTI val Car AP3D Mod. | AP3D: 74.65 -> 76.71 | Improved |
| Use GIoU loss instead of IoU loss for 2D-3D constraint (ablation) | Output-level guidance · KITTI val Car AP3D Mod. | AP3D: 74.50 -> 76.71 | Improved |
| Use GIoU loss instead of L1 corner regression (ablation) | Output-level guidance · KITTI val Car AP3D Mod. | AP3D: 74.70 -> 76.71 | Improved |
| Add 2D-box overlap filter and score filter to pseudo-labeling instead of plain pseudo-labeling (ablation) | Training-level guidance · KITTI val Car AP3D Mod. | AP3D: 76.64 -> 78.89 | Improved |
| Add overlap filter only to pseudo-labeling (ablation) | Training-level guidance · KITTI val Car AP3D Mod. | AP3D: 76.64 -> 77.55 | Improved |
| Run more pseudo-label self-training rounds (ablation) | Training-level guidance · KITTI val pseudo-label recall IoU=0.7 | recall: 0.4671 -> 0.7422 after round 2, 0.7421 round 3 | Improved |
Closest papers by meaning
- Weakly Supervised 3D Object Detection from Lidar Point Cloud (ECCV 2020 · similarity 0.93)
- WeakM3D: Towards Weakly Supervised Monocular 3D Object Detection (ICLR 2022 · similarity 0.93)
- VSRD: Instance-Aware Volumetric Silhouette Rendering for Weakly Supervised 3D Object Detection (CVPR 2024 · similarity 0.91)
- Leveraging Imagery Data with Spatial Point Prior for Weakly Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.90)
- LabelAny3D: Label Any Object 3D in the Wild (NeurIPS 2025 · similarity 0.89)
- Improving Distant 3D Object Detection Using 2D Box Supervision (CVPR 2024 · 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.
