Research map / Detection and segmentation / 3D detection and driving
MotionNet: Joint Perception and Motion Prediction for Autonomous Driving Based on Bird's Eye View Maps
Pengxiang Wu, Siheng Chen, Dimitris N. Metaxas
Check 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 MotionNet instead of LSTM-Encoder-Decoder | nuScenes environmental state prediction · nuScenes test | mean displacement error, speed > 5m/s: 1.5885 -> 1.0744 | Improved |
| Use MotionNet instead of LSTM-Encoder-Decoder | nuScenes environmental state prediction · nuScenes test | inference time: 0.042s -> 0.019s | Improved |
| Add spatial consistency loss L_s (ablation) | MotionNet · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2565 -> 0.2488 | Improved |
| Add foreground temporal consistency loss L_ft (ablation) | MotionNet · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2565 -> 0.2515 | Improved |
| Add background temporal consistency loss L_bt (ablation) | MotionNet · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2565 -> 0.2530 | Improved |
| Add all three consistency losses (ablation) | MotionNet · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2565 -> 0.2467 | Improved |
| Use MGDA for multi-task optimization (ablation) | MotionNet · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2565 -> 0.2366 | Improved |
| Use MGDA alongside all consistency losses (ablation) | MotionNet · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2565 -> 0.2292 | Improved |
| Increase input history from 2 to 5 frames (ablation) | MotionNet · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2921 -> 0.2565 | Improved |
| Remove ego-motion synchronization (ablation) | MotionNet · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2565 -> 0.4245 | Hurt |
| Replace ground-truth sweep synchronization with ICP (ablation) | MotionNet · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2565 -> 0.4073 | Hurt |
| Use middle temporal fusion rather than early fusion (ablation) | MotionNet with STC blocks · nuScenes test | mean displacement error, speed ≤ 5m/s: 0.2596 -> 0.2565 | Improved |
Later papers that compared with this method
1 later accepted paper reports MotionNet as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Weakly Supervised Class-Agnostic Motion Prediction for Autonomous Driving (CVPR 2023 · better on 1 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
- Sequential Multi-View Fusion Network for Fast LiDAR Point Motion Estimation (ECCV 2022 · similarity 0.89)
- Argoverse 2: Next Generation Datasets for Self-Driving Perception and Forecasting (NeurIPS 2021 · similarity 0.89)
- PriorMotion: Generative Class-Agnostic Motion Prediction with Raster-Vector Motion Field Priors (ICCV 2025 · similarity 0.88)
- PnPNet: End-to-End Perception and Prediction With Tracking in the Loop (CVPR 2020 · similarity 0.88)
- Self-Supervised Class-Agnostic Motion Prediction with Spatial and Temporal Consistency Regularizations (CVPR 2024 · similarity 0.87)
- Self-Supervised Bird’s Eye View Motion Prediction with Cross-Modality Signals (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.
