Research map / Detection and segmentation / 3D detection and driving
RandomRooms: Unsupervised Pre-Training From Synthetic Shapes and Randomized Layouts for 3D Object Detection
Yongming Rao, Benlin Liu, Yi Wei, Jiwen Lu, Cho-Jui Hsieh, Jie Zhou
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 RandomRooms pre-training | VoteNet · ScanNetV2 val | mAP@0.25: 58.6 -> 61.3 | Improved |
| add RandomRooms pre-training | H3DNet · ScanNetV2 val | mAP@0.5: 48.1 -> 51.5 | Improved |
| add RandomRooms pre-training | VoteNet · SUN RGB-D val | mAP@0.5: 32.9 -> 35.4 | Improved |
| add RandomRooms pre-training | H3DNet · SUN RGB-D val | mAP@0.25: 60.1 -> 61.6 | Improved |
| reduce training data to 5% with RandomRooms pre-training | VoteNet · ScanNetV2 val (5% data) | mAP@0.25: 12.6 -> 19.1 | Improved |
| reduce training data to 5% with RandomRooms pre-training | H3DNet · ScanNetV2 val (5% data) | mAP@0.25: 26.6 -> 32.0 | Improved |
| pre-train on ShapeNet instead of ScanNetV2 (ablation) | VoteNet RandomRooms · ScanNetV2 val | mAP@0.25: 60.2 -> 61.3 | Improved |
| pre-train on both ShapeNet and ScanNetV2 (ablation) | VoteNet RandomRooms · ScanNetV2 val | mAP@0.25: 61.3 -> 61.5 | Improved |
| use instance-level instead of point-level contrastive loss (ablation) | VoteNet pre-training · ScanNetV2 val | mAP@0.25: 59.2 -> 61.3 | Improved |
| add segmentation loss to instance-level contrastive (ablation) | VoteNet pre-training · ScanNetV2 val | mAP@0.25: 61.3 -> 61.5 | No clear effect |
| add gravity principle in layout generation (ablation) | RandomRooms · ScanNetV2 val | mAP@0.25: 60.5 -> 61.3 | Improved |
| add floor and wall to generated scene (ablation) | RandomRooms · ScanNetV2 val | mAP@0.25: 60.7 -> 61.3 | 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.
- H3DNet — H3DNet: 3D Object Detection Using Hybrid Geometric Primitives (ECCV 2020) · better on 6 of 6 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
- Self-Supervised Pretraining of 3D Features on Any Point-Cloud (ICCV 2021 · similarity 0.88)
- Finding Your (3D) Center: 3D Object Detection Using a Learned Loss (ECCV 2020 · similarity 0.88)
- Prompt3D: Random Prompt Assisted Weakly-Supervised 3D Object Detection (CVPR 2024 · similarity 0.88)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.88)
- Cubify Anything: Scaling Indoor 3D Object Detection (CVPR 2025 · similarity 0.87)
- MonoPlace3D: Learning 3D-Aware Object Placement for 3D Monocular Detection (CVPR 2025 · similarity 0.87)
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.
