Research map / Point clouds / Segmentation and detection
Multimodal Transformer for Automatic 3D Annotation and Object Detection
Chang Liu, Xiaoyan Qian, Binxiao Huang, Xiaojuan Qi, Edmund Lam, Siew-Chong Tan, Ngai Wong
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 image modality (ablation) | vanilla LiDAR transformer · KITTI val box quality (500 frames) | mIoU: 60.07 -> 65.57 | Improved |
| add image modality (ablation) | vanilla LiDAR transformer · KITTI val box quality (500 frames) | Recall: 41.53 -> 53.30 | Improved |
| add segmentation loss (ablation) | Multimodal variant · KITTI val box quality (500 frames) | mAP R40: 57.33 -> 69.42 | Improved |
| add point generation (XYZ) loss (ablation) | Multimodal variant · KITTI val box quality (500 frames) | mAP R40: 57.33 -> 63.82 | Improved |
| add self-supervision on unlabeled frames (ablation) | XYZ Loss variant · KITTI val box quality (500 frames) | mIoU: 67.61 -> 68.42 | Improved |
| add extra target points sampled from image (ablation) | XYZ Loss variant · KITTI val box quality (500 frames) | mAP: 63.62 -> 80.82 | Improved |
| add self-supervision on top of Densify (ablation) | Densify variant · KITTI val box quality (500 frames) | mIoU: 75.37 -> 76.84 | Improved |
| add segmentation loss on top of Full Generate (ablation) | Full Generate · KITTI val box quality (500 frames) | mAP R40: 85.72 -> 90.30 | Improved |
| train with all frames instead of 500 (ablation) | MTrans · KITTI val box quality | mAP: 86.24 -> 92.36 | Improved |
| retrain PointPillars on MTrans labels (500 frames) | PointPillars · KITTI val | AP3D Moderate: 67.65 -> 76.56 | Improved |
| retrain PointPillars on MTrans labels (125 frames) | PointPillars · KITTI val | AP3D Moderate: 58.29 -> 71.66 | Improved |
| retrain PointRCNN on MTrans labels (125 frames) | PointRCNN · KITTI val | AP3D Moderate: 73.60 -> 77.31 | Improved |
Closest papers by meaning
- MsSVT: Mixed-scale Sparse Voxel Transformer for 3D Object Detection on Point Clouds (NeurIPS 2022 · similarity 0.89)
- A Simple Vision Transformer for Weakly Semi-supervised 3D Object Detection (ICCV 2023 · similarity 0.89)
- Detecting As Labeling: Rethinking LiDAR-camera Fusion in 3D Object Detection (ECCV 2024 · similarity 0.89)
- iDet3D: Towards Efficient Interactive Object Detection for LiDAR Point Clouds (AAAI 2024 · similarity 0.88)
- STONE: A Submodular Optimization Framework for Active 3D Object Detection (NeurIPS 2024 · similarity 0.88)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · 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.
