Research map / Label-efficient and robust learning / Self-supervised and contrastive
Differentiable Raycasting for Self-Supervised Occupancy Forecasting
Tarasha Khurana, Peiyun Hu, Achal Dave, Jason Ziglar, David Held, Deva Ramanan
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 differentiable raycasting to decouple scene and ego motion | freespace forecaster · nuScenes val | F1: 0.665 -> 0.777 | Improved |
| add differentiable raycasting to decouple scene and ego motion | freespace forecaster · ONCE val | F1: 0.635 -> 0.787 | Improved |
| add differentiable raycasting (ablation) | motion planner · nuScenes val | box collision 3s (%): 1.10 -> 0.88 | Improved |
| add differentiable raycasting (ablation) | motion planner · nuScenes val | L2 error 3s (m): 2.64 -> 2.78 | Hurt |
| add multi-task occupancy forecasting auxiliary task (ablation) | IL max-margin planner · nuScenes val | box collision 3s (%): 1.95 -> 1.29 | Improved |
| add multi-task occupancy forecasting auxiliary task (ablation) | IL max-margin planner · nuScenes val | L2 error 3s (m): 2.47 -> 2.30 | No clear effect |
| add freespace-guided cost margin (ablation) | IL max-margin planner · nuScenes val | box collision 3s (%): 1.95 -> 1.07 | Improved |
| add freespace-guided cost margin (ablation) | IL max-margin planner · nuScenes val | L2 error 3s (m): 2.47 -> 2.54 | Hurt |
| add freespace-guided cost margin to multi-task planner (ablation) | multi-task planner · nuScenes val | box collision 3s (%): 1.29 -> 1.10 | Improved |
| add differentiable raycasting (ablation) | motion planner · ONCE val | box collision 3s (%): 2.61 -> 2.79 | Hurt |
| add differentiable raycasting (ablation) | motion planner · ONCE val | point collision 3s (%): 1.10 -> 0.74 | Improved |
| increase self-supervised training data from 8K to 94K | Ours · ONCE val | L2 error 3s (m): 4.43 -> 2.90 | Improved |
Closest papers by meaning
- UnO: Unsupervised Occupancy Fields for Perception and Forecasting (CVPR 2024 · similarity 0.91)
- COME: Adding Scene-Centric Forecasting Control to Occupancy World Model (NeurIPS 2025 · similarity 0.90)
- Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · similarity 0.90)
- Semi-Supervised Vision-Centric 3D Occupancy World Model for Autonomous Driving (ICLR 2025 · similarity 0.89)
- OccProphet: Pushing the Efficiency Frontier of Camera-Only 4D Occupancy Forecasting with an Observer-Forecaster-Refiner Framework (ICLR 2025 · similarity 0.88)
- Spatiotemporal Decoupling for Efficient Vision-Based Occupancy Forecasting (CVPR 2025 · 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.
