Research map / Point clouds / Segmentation and detection
MSF: Motion-Guided Sequential Fusion for Efficient 3D Object Detection From Point Cloud Sequences
Chenhang He, Ruihuang Li, Yabin Zhang, Shuai Li, Lei Zhang
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 |
|---|---|---|---|
| propagate current-frame proposals using estimated motion and fuse sequential points | MSF · Waymo Open Dataset test | 3D mAPH L1: 80.91 -> 81.74 | Improved |
| add Motion Embedding (ablation) | MSF · Waymo Open Dataset validation | APH L1: 80.03 -> 80.20; +0.17 | Improved |
| add Motion Embedding (ablation) | MSF · Waymo Open Dataset validation | APH L2: 74.50 -> 74.62; +0.12 | Improved |
| add Self-Attention for intra-proposal spatial interactions (ablation) | MSF · Waymo Open Dataset validation | APH L1: 76.51 -> 80.20; +3.69 | Improved |
| add Self-Attention for intra-proposal spatial interactions (ablation) | MSF · Waymo Open Dataset validation | APH L2: 71.91 -> 74.62; +2.71 | Improved |
| add BiFA cross-frame aggregation (ablation) | MSF · Waymo Open Dataset validation | APH L1: 78.25 -> 80.20; +1.95 | Improved |
| add BiFA cross-frame aggregation (ablation) | MSF · Waymo Open Dataset validation | APH L2: 73.11 -> 74.62; +1.51 | Improved |
| use both forward and backward aggregation paths rather than forward-only aggregation (ablation) | MSF · Waymo Open Dataset validation | APH L1: 79.64 -> 80.20; +0.56 | Improved |
| use both forward and backward aggregation paths rather than backward-only aggregation (ablation) | MSF · Waymo Open Dataset validation | APH L2: 73.77 -> 74.62; +0.85 | Improved |
| use a query-based decoder instead of single max-pooling (ablation) | MSF · Waymo Open Dataset validation | APH L1: 79.54 -> 80.20; +0.66 | Improved |
| replace raw points with proxy points (ablation) | MSF with Self-Attention · Waymo Open Dataset validation, Pedestrian L2 | APH: 75.11 -> 74.20; -0.91 | Hurt |
| replace raw points and Self-Attention with proxy points and MLP Mixer (ablation) | MSF · Waymo Open Dataset validation, Vehicle L2 | APH: 73.35 -> 73.45; +0.10 | 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.
- CenterFormer — CenterFormer: Center-based Transformer for 3D Object Detection (ECCV 2022) · better on 2 of 2 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
- MPPNet: Multi-Frame Feature Intertwining with Proxy Points for 3D Temporal Object Detection (ECCV 2022 · similarity 0.89)
- Homogeneous Multi-modal Feature Fusion and Interaction for 3D Object Detection (ECCV 2022 · similarity 0.88)
- Structure Aware Single-Stage 3D Object Detection From Point Cloud (CVPR 2020 · similarity 0.87)
- PointDistiller: Structured Knowledge Distillation Towards Efficient and Compact 3D Detection (CVPR 2023 · similarity 0.87)
- MoDAR: Using Motion Forecasting for 3D Object Detection in Point Cloud Sequences (CVPR 2023 · similarity 0.87)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 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.
