Research map / Label-efficient and robust learning / Self-supervised and contrastive
Forecast-MAE: Self-supervised Pre-training for Motion Forecasting with Masked Autoencoders
Jie Cheng, Xiaodong Mei, Ming Liu
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 agents' future trajectories as input to lane-masked pre-training (ablation) | Forecast-MAE · Argoverse 2 val | minADE6: 0.865 -> 0.828 | Improved |
| mask history only with future given unmasked (ablation) | Forecast-MAE · Argoverse 2 val | minADE6: 0.8314 -> 0.864 | Hurt |
| add future masking (complementary with history) (ablation) | Forecast-MAE · Argoverse 2 val | minADE6: 0.866 -> 0.814 | Improved |
| increase encoder depth from 2 to 4 (ablation) | Forecast-MAE · Argoverse 2 val | minADE6: 0.854 -> 0.814 | Improved |
| increase encoder depth from 4 to 5 (ablation) | Forecast-MAE · Argoverse 2 val | minADE6: 0.814 -> 0.815 | No clear effect |
| add self-supervised pre-training before fine-tuning | Forecast-MAE · Argoverse 2 val | minADE6: 0.811 -> 0.801 | Improved |
| train from scratch for more epochs (60 to 80) (ablation) | Forecast-MAE scratch · Argoverse 2 val | minADE6: 0.811 -> 0.814 | No clear effect |
| add pre-training under distribution shift between city groups (ablation) | Forecast-MAE · Argoverse 2 (train Miami/Pittsburgh/Austin, test other cities) | minADE: 0.9098 -> 0.8968 | Improved |
| add MAE pre-training vs SSL-Lanes pretext tasks | Forecast-MAE vs SSL-Lanes · Argoverse 2 val | minADE6: 0.850 -> 0.801 | Improved |
| add ensemble of 6 variants (ablation) | Forecast-MAE · Argoverse 2 test | minFDE1: 4.355 -> 4.145 | Improved |
| add pre-training on Waymo Open Motion Dataset | Forecast-MAE · WOMD test | minADE: 0.689 -> 0.632 | Improved |
| add pre-training on Argoverse 1 | Forecast-MAE · Argoverse 1 val | minADE: 0.736 -> 0.710 | Improved |
Closest papers by meaning
- SEPT: Towards Efficient Scene Representation Learning for Motion Prediction (ICLR 2024 · similarity 0.88)
- SmartPretrain: Model-Agnostic and Dataset-Agnostic Representation Learning for Motion Prediction (ICLR 2025 · similarity 0.87)
- Not All Data are Good Labels: On the Self-supervised Labeling for Time Series Forecasting (NeurIPS 2025 · similarity 0.85)
- VisionTS: Visual Masked Autoencoders Are Free-Lunch Zero-Shot Time Series Forecasters (ICML 2025 · similarity 0.85)
- T4P: Test-Time Training of Trajectory Prediction via Masked Autoencoder and Actor-specific Token Memory (CVPR 2024 · similarity 0.85)
- Self-Supervised Class-Agnostic Motion Prediction with Spatial and Temporal Consistency Regularizations (CVPR 2024 · similarity 0.85)
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.
