Research map / Detection and segmentation / Weakly supervised detection and segmentation
Learning from Noisy Data for Semi-Supervised 3D Object Detection
Zehui Chen, Zhenyu Li, Shuo Wang, Dengpan Fu, Feng Zhao
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 noise-resistant (soft) classification supervision (ablation) | NoisyStudent baseline · ONCE val (Small) | mAP: 55.50 -> 56.80 | Improved |
| add Gaussian NLL regression on top of classification supervision (ablation) | NoiseDet · ONCE val (Small) | mAP: 56.80 -> 57.27 | Improved |
| add dense feature consistency constraint (ablation) | NoiseDet · ONCE val (Small) | mAP: 57.27 -> 58.01 | Improved |
| soft quality-based targets instead of Focal Loss on hard pseudo labels (ablation) | classification supervision · ONCE val (Small) | AP overall: 55.45 -> 56.80 | Improved |
| combine IoU and confidence by weighted sum instead of multiplication (ablation) | classification target · ONCE val (Small) | AP overall: 56.52 -> 56.80 | Improved |
| feature-level L2 consistency instead of label-level L2 (ablation) | consistency regularization · ONCE val (Small) | AP: 55.45 -> 56.70 | Improved |
| feature-level distance correlation instead of plain L2 (ablation) | consistency regularization · ONCE val (Small) | AP: 56.70 -> 56.07 | Hurt |
| Gaussian NLL regression loss instead of Smooth L1 or L1 (ablation) | regression supervision · ONCE val (Small) | mAP: 56.4 mAP | Improved |
| replace NoisyStudent with NoiseDet on SECOND | SECOND · ONCE val (Small) | mAP: 55.50 -> 58.01 | Improved |
| replace NoisyStudent with NoiseDet on CenterPoint | CenterPoint · ONCE val (Small) | mAP: 63.77 -> 65.57 | Improved |
| use higher-quality pseudo labels from CenterPoint-E teacher | NoiseDet with SECOND student · ONCE val | mAP: 58.00 -> 60.19 | Improved |
| increase amount of unlabeled data from Small to Large | NoiseDet · ONCE val | mAP: 58.00 -> 61.16 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- 3DIoUMatch — 3DIoUMatch: Leveraging IoU Prediction for Semi-Supervised 3D Object Detection (CVPR 2021) · better on 2 of 2 shared results
- SESS — SESS: Self-Ensembling Semi-Supervised 3D Object Detection (CVPR 2020) · better on 1 of 1 shared result
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
- Adapting Object Size Variance and Class Imbalance for Semi-supervised Object Detection (AAAI 2023 · similarity 0.91)
- PseCo: Pseudo Labeling and Consistency Training for Semi-Supervised Object Detection (ECCV 2022 · similarity 0.91)
- Diffusion-SS3D: Diffusion Model for Semi-supervised 3D Object Detection (NeurIPS 2023 · similarity 0.91)
- Rethinking Pseudo Labels for Semi-supervised Object Detection (AAAI 2022 · similarity 0.91)
- Decoupled Pseudo-labeling for Semi-Supervised Monocular 3D Object Detection (CVPR 2024 · similarity 0.90)
- Leveraging Temporal Cues for Semi-Supervised Multi-View 3D Object Detection (CVPR 2025 · similarity 0.90)
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.
