Research map / Point clouds / Point cloud classification
Sequential Point Clouds: A Survey
Haiyan Wang, Yingli Tian
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 multiple frames instead of a single frame | MVF++ · Waymo Open val Vehicles (IoU=0.7) | 3D AP: 74.64 -> 79.73 | Improved |
| use offboard multi-frame detection instead of single-frame MVF++ | Qi et al. · Waymo Open val Vehicles (IoU=0.7) | 3D AP: 74.64 -> 84.50 | Improved |
| use graph-based multi-frame detector instead of best single-frame | Yin et al. · nuScenes detection | mean AP: 42.0 -> 45.4 | Improved |
| use convolution-based multi-frame detector instead of single-frame PointPillars | What you see · nuScenes detection | mean AP: 30.5 -> 35.0 | Improved |
| use point-based network instead of voxel-based network for scene flow | FlowNet3D vs VoxFlowNet · FlyingThings3D scene flow | EPE (m): 0.2971 -> 0.1694 | Improved |
| train scene flow without supervision | PointPWCNet · FlyingThings3D scene flow | EPE (m): 0.0588 -> 0.1246 | Hurt |
| train on FlyingThings3D and test on real KITTI without retraining | PV-RAFT · KITTI scene flow | EPE (m): 0.0461 -> 0.056 | No clear effect |
| use 4 scans instead of 1 scan as input | Aygun et al. · SemanticKITTI 4D panoptic val | LSTQ: 43.88 -> 56.89 | Improved |
| propagate instance IDs with scene flow instead of multi-object tracking | Aygun et al. · SemanticKITTI 4D panoptic val | LSTQ: 40.18 -> 43.88 | Improved |
| use multiple frames instead of one frame with sparse voxel network | 4D MinkNet14 · Synthia 4D semantic segmentation | mIoU: 76.24 -> 77.46 | Improved |
| increase PSTNet temporal kernel l from 1 to 3 | PSTNet · Synthia 4D semantic segmentation | mIoU: 80.79 -> 82.24 | Improved |
| use RNN-based instead of convolution/graph-based detectors for temporal modeling | RNN-based detectors · nuScenes detection | mean AP | Hurt |
Closest papers by meaning
- Deep Learning for 3D Point Clouds: A Survey (TPAMI 2020 · similarity 0.89)
- Learning Scene Dynamics from Point Cloud Sequences (IJCV 2022 · similarity 0.88)
- S2Net: Stochastic Sequential Pointcloud Forecasting (ECCV 2022 · similarity 0.86)
- MSF: Motion-Guided Sequential Fusion for Efficient 3D Object Detection From Point Cloud Sequences (CVPR 2023 · similarity 0.86)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.86)
- Dynamic 3D Point Cloud Sequences as 2D Videos (TPAMI 2024 · similarity 0.85)
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.
