Research map / Point clouds / Segmentation and detection
MGTANet: Encoding Sequential LiDAR Points Using Long Short-Term Motion-Guided Temporal Attention for 3D Object Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply MGTANet to CenterPoint with test-time augmentation | CenterPoint · nuScenes test | mAP: 60.3 -> 67.5 | Helped |
| Add short-term motion-aware voxel feature encoding (ablation) | CenterPoint · nuScenes val (1/7 training set) | mAP: 54.99 -> 56.52 | Helped |
| Add spatio-temporal feature aggregation without feature alignment (ablation) | CenterPoint with SM-VFE · nuScenes val (1/7 training set) | mAP: 56.52 -> 58.24 | Helped |
| Add motion-guided deformable alignment (ablation) | CenterPoint with SM-VFE and STFA · nuScenes val (1/7 training set) | mAP: +1.37% | Helped |
| Add motion embedding to voxel features (ablation) | CenterPoint · nuScenes val (1/7 training set) | mAP: 54.99 -> 56.24 | Helped |
| Add channel-wise attention to motion embedding (ablation) | CenterPoint with motion embedding · nuScenes val (1/7 training set) | mAP: +0.28% | Helped |
| Use derivative queries instead of a single query (ablation) | STFA · nuScenes val (1/7 training set) | mAP: 57.50 -> 58.24 | Helped |
| Remove motion-guided deformable alignment (ablation) | CenterPoint with feature aggregation · nuScenes val (1/7 training set) | mAP: 54.99 -> 52.63 | Hurt |
| Add motion-guided deformable alignment rather than simple concatenation and convolution (ablation) | CenterPoint with feature aggregation · nuScenes val (1/7 training set) | mAP: 54.99 -> 58.37 | Helped |
| Use offline mode with a previous frame and current frame | MGTANet-C · nuScenes val (full training set) | mAP: 62.9 -> 64.8 | Helped |
Closest papers by meaning
- PTT: Point-Trajectory Transformer for Efficient Temporal 3D Object Detection (CVPR 2024 · similarity 0.92)
- Not All Points Are Equal: Learning Highly Efficient Point-Based Detectors for 3D LiDAR Point Clouds (CVPR 2022 · similarity 0.91)
- iDet3D: Towards Efficient Interactive Object Detection for LiDAR Point Clouds (AAAI 2024 · similarity 0.89)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 2024 · similarity 0.89)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.89)
- BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024 · similarity 0.88)
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.