Research map / Label-efficient and robust learning / Self-supervised and contrastive
Weakly and Self-Supervised Class-Agnostic Motion Prediction for Autonomous Driving
Ruibo Li, Hanyu Shi, Zhe Wang, Guosheng Lin
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 Geman-McClure penalty instead of L2 in Chamfer loss (ablation) | WeakMotionNet · nuScenes WeakMotion-FB 1% | mean error, speed > 5m/s: 2.274 -> 1.662 | Improved |
| add past-frame point clouds as targets (ablation) | Chamfer-GM loss · nuScenes WeakMotion-FB 1% | mean error, speed > 5m/s: 1.66m to 1.46m | Improved |
| add multi-frame consistency reweighting (ablation) | multi-frame Chamfer-GM · nuScenes WeakMotion-FB 1% | mean error: static 0.083 -> 0.060; fast 1.464 -> 1.574 | Mixed |
| add smoothness regularization to RCCD loss (ablation) | RCCD loss · nuScenes WeakMotion-FB 1% | mean error, speed > 5m/s: 1.574 -> 1.134 | Improved |
| add auxiliary FG/BG segmentation head (ablation) | RCCD + Smooth · nuScenes WeakMotion-FB 1% | mean error: static 0.039 -> 0.023; fast 1.134 -> 1.211 | Mixed |
| derive supervision from predicted foreground points instead of raw points (ablation) | WeakMotion-FB · nuScenes (1% FG/BG masks) | mean error, speed > 5m/s: 6.093 -> 1.211 | Improved |
| derive supervision from predicted non-ground points instead of raw points (ablation) | WeakMotion-NG · nuScenes (0.01% FG/BG masks) | mean error, speed > 5m/s: 6.363 -> 1.642 | Improved |
| derive supervision from non-ground points instead of raw points (ablation) | SelfMotion-NG · nuScenes (self-supervised) | mean error, static: 1.721 -> 0.044 | Improved |
| use predicted foreground instead of non-ground points for motion supervision (ablation) | WeakMotion · nuScenes (1% vs 0.01% FG/BG masks) | mean error, speed > 5m/s: 1% masks: 1.533 -> 1.211; 0.01% masks: 1.642 -> 1.810 | Mixed |
| reduce FG/BG masks from 100% to 1% (ablation) | WeakMotion-FB · nuScenes test | mean error, speed > 5m/s: 1.1974 -> 1.2108 | No clear effect |
| reduce PreSegNet masks from 1% to 0.01% (ablation) | PreSegNet · nuScenes val | overall accuracy: 90.0% -> 85.3% | Hurt |
| RCCD loss and smoothness replace CCD loss | WeakMotion-FB (1%) · nuScenes test | mean error, speed > 5m/s: 1.7823 -> 1.2108 | Improved |
Closest papers by meaning
- Self-Supervised Class-Agnostic Motion Prediction with Spatial and Temporal Consistency Regularizations (CVPR 2024 · similarity 0.94)
- Semi-supervised Class-Agnostic Motion Prediction with Pseudo Label Regeneration and BEVMix (AAAI 2024 · similarity 0.87)
- Self-Supervised Bird’s Eye View Motion Prediction with Cross-Modality Signals (AAAI 2024 · similarity 0.87)
- UnO: Unsupervised Occupancy Fields for Perception and Forecasting (CVPR 2024 · similarity 0.87)
- VoteFlow: Enforcing Local Rigidity in Self-Supervised Scene Flow (CVPR 2025 · similarity 0.86)
- SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction (ICLR 2025 · 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.
