Research map / Point clouds / Point cloud classification
MeteorNet: Deep Learning on Dynamic 3D Point Cloud Sequences
Xingyu Liu, Mengyuan Yan, Jeannette Bohg
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 more input frames to MeteorNet-cls (4 -> 24) (ablation) | MeteorNet-cls · MSRAction3D | accuracy: 78.11 -> 88.50 | Improved |
| use a multi-frame point cloud sequence (24 frames) instead of a single frame | PointNet++ · MSRAction3D | accuracy: 61.61 -> 88.50 | Improved |
| increase frames from 16 to 24 for action recognition (ablation) | MeteorNet-cls · MSRAction3D | accuracy: 88.21 -> 88.50 | No clear effect |
| add multiple frames to a single-frame point network | PointNet++ · Synthia (segmentation) | mIoU: 79.35 -> 81.80 | Improved |
| enlarge the max-pooling bottleneck dimension (seg-s to seg-m) (ablation) | MeteorNet-seg · Synthia (segmentation) | mIoU: 80.31 -> 81.72 | Improved |
| enlarge non-bottleneck layer widths (seg-m to seg-l) with 3 frames (ablation) | MeteorNet-seg · Synthia (segmentation) | mIoU: 81.13 -> 81.80 | Improved |
| increase input frames from 2 to 3 without enlarging the model (ablation) | MeteorNet-seg-m · Synthia (segmentation) | mIoU: 81.72 -> 81.13 | Hurt |
| use point-based MeteorNet instead of sparse-convolution MinkNet | 4D MinkNet14 · Synthia (segmentation) | mIoU: 77.46 -> 81.80 | Improved |
| add a third input frame (ablation) | MeteorNet-seg · KITTI (segmentation) | mIoU: 71.65 -> 73.42 | Improved |
| add a third frame for the Car class (ablation) | MeteorNet-seg · KITTI (segmentation, Car IoU) | IoU: 74.53 -> 71.22 | Hurt |
| add a fourth frame for scene flow (chained-flow) (ablation) | MeteorNet-flow · FlyingThings3D (scene flow) | mean EPE: 0.215 -> 0.209 | Improved |
| use chained-flow grouping instead of direct grouping (4 frames) (ablation) | MeteorNet-flow · FlyingThings3D (scene flow) | accuracy (EPE<0.1): 52.12 -> 49.91 | Hurt |
Later papers that compared with this method
5 later accepted papers report Meteornet as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.
- FlowCloud: Learning Continuous Spatiotemporal Dynamics from Unpaired Sparse Point Cloud Snapshots (ICML 2026 · better on 2 of 2 shared results)
- Point Contrastive Prediction with Semantic Clustering for Self-Supervised Learning on Point Cloud Videos (ICCV 2023 · better on 1 of 1 shared result)
- Point Primitive Transformer for Long-Term 4D Point Cloud Video Understanding (ECCV 2022 · better on 2 of 2 shared results)
- PSTNet: Point Spatio-Temporal Convolution on Point Cloud Sequences (ICLR 2021 · better on 8 of 8 shared results)
- CGA-Net: Category Guided Aggregation for Point Cloud Semantic Segmentation (CVPR 2021 · 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
- Dynamic 3D Point Cloud Sequences as 2D Videos (TPAMI 2024 · similarity 0.86)
- Fully-Convolutional Point Networks for Large-Scale Point Clouds (ECCV 2018 · similarity 0.86)
- PSTNet: Point Spatio-Temporal Convolution on Point Cloud Sequences (ICLR 2021 · similarity 0.86)
- PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation (CVPR 2017 · similarity 0.86)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.86)
- FlowNet3D: Learning Scene Flow in 3D Point Clouds (CVPR 2019 · similarity 0.86)
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.
