Research map / Detection and segmentation / 3D detection and driving
UpCycling: Semi-supervised 3D Object Detection without Sharing Raw-level Unlabeled Scenes
Sunwook Hwang, Youngseok Kim, Seongwon Kim, Saewoong Bahk, Hyung-Sin Kim
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 |
|---|---|---|---|
| use feature-level ground-truth sampling instead of feature-level flip (ablation) | UpCycling · KITTI partial-label (10% labeled), SECOND-IoU | AP3D Mod gain over baseline: -20.09 -> +7.87 | Improved |
| add feature-level noise or random sampling instead of GT sampling (ablation) | UpCycling · KITTI partial-label (10% labeled), SECOND-IoU | AP3D: +0.03 / +0.13 / -1.23 (Noise); +2.47 / -0.96 / +0.63 (RS) | No clear effect |
| combine feature-level Flip and RS (F-3DIoUMatch) (ablation) | 3DIoUMatch at feature level · KITTI partial-label (10% labeled), SECOND-IoU | AP3D: -11.69 / -13.75 / -13.32 | Hurt |
| use feature-level GT sampling instead of feature-level rotation (ablation) | UpCycling · KITTI partial-label (10% labeled), SECOND-IoU | AP3D Hard gain over baseline: +7.96 -> +8.14 | Improved |
| apply UpCycling on features without Statistical Normalization for Waymo-to-KITTI adaptation | UpCycling · Waymo to KITTI domain adaptation | AP3D: 11.71 vs ST3D 40.16 (SECOND-IoU) | Hurt |
| combine UpCycling with statistical normalization (SN) | UpCycling · Waymo to KITTI domain adaptation | AP3D: 11.71 -> 67.65 (SECOND-IoU) | Improved |
| UpCycling versus ST3D on Waymo to Lyft | UpCycling · Waymo to Lyft domain adaptation | AP3D: 29.90 -> 45.66 (SECOND-IoU) | Improved |
| UpCycling versus 3DIoUMatch with 25% labels | UpCycling · KITTI partial-label (25% labeled) | AP3D | Improved |
| train the backbone with unlabeled raw data (raw-level UpCycling) instead of freezing it (ablation) | UpCycling · KITTI partial-label | AP3D | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ST3D — ST3D: Self-Training for Unsupervised Domain Adaptation on 3D Object Detection (CVPR 2021) · better on 3 of 3 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
- Leveraging Temporal Cues for Semi-Supervised Multi-View 3D Object Detection (CVPR 2025 · similarity 0.91)
- Semi-Supervised 3D Object Detection with Proficient Teachers (ECCV 2022 · similarity 0.91)
- DetMatch: Two Teachers Are Better than One for Joint 2D and 3D Semi-Supervised Object Detection (ECCV 2022 · similarity 0.90)
- Dual-Perspective Knowledge Enrichment for Semi-supervised 3D Object Detection (AAAI 2024 · similarity 0.89)
- Semi-supervised 3D Object Detection with PatchTeacher and PillarMix (AAAI 2024 · similarity 0.89)
- Decoupled Pseudo-labeling for Semi-Supervised Monocular 3D Object Detection (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.
