Research map / Multi-agent systems and games / LLM agents
OccDirector: Language-Guided Behavior and Interaction Generation in 4D Occupancy Space
Zhuding Liang, Tianyi Yan, Dubing Chen, Jiasen Zheng, Huan Zheng, Cheng-zhong Xu, Yida Wang, Kun Zhan, Shen Jianbing
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 history-prefix anchoring (ablation) | OccDirector · OccInteract-85k video | FVD: 354 -> 275 | Improved |
| add history-prefix anchoring (ablation) | OccDirector · OccInteract-85k video | VLM instruction alignment mean: 3.43 -> 3.73 | Improved |
| replace full attention with STSA (ablation) | OccDirector · OccInteract-85k ablation | FVD: 532 -> 354 | Improved |
| replace full attention with STSA (ablation) | OccDirector · OccInteract-85k ablation | VLM mean score: 2.78 -> 3.43 | Improved |
| add Token Refiner with STSA (ablation) | OccDirector · OccInteract-85k ablation | FVD: 407 -> 354 | Improved |
| add Token Refiner with full attention (ablation) | OccDirector · OccInteract-85k ablation | VLM mean score: 2.77 -> 2.78 | No clear effect |
| use VLM encoder with MMDiT instead of DynamicCity + CLIP text | DynamicCity + CLIP · OccInteract-85k video | FVD: 388 -> 275 | Improved |
| use VLM text encoder instead of CLIP/T5 | DynamicCity text baselines · OccInteract-85k video | video Recall: 0.008 -> 0.12 | Improved |
| use OccDirector without history control instead of DynamicCity + CLIP | DynamicCity + CLIP · OccInteract-85k video | video Precision: 0.60 -> 0.48 | Hurt |
| use text-guided OccDirector instead of geometry-conditioned DynamicCity | DynamicCity · OccInteract-85k single-frame | FID: 61 -> 62 | No clear effect |
| use OccDirector instead of DynamicCity + CLIP | DynamicCity + CLIP · OccInteract-85k single-frame | FID: 65 -> 62 | Improved |
Closest papers by meaning
- Driving in the Occupancy World: Vision-Centric 4D Occupancy Forecasting and Planning via World Models for Autonomous Driving (AAAI 2025 · similarity 0.90)
- UniScene: Unified Occupancy-centric Driving Scene Generation (CVPR 2025 · similarity 0.86)
- DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving Scene Representation (CVPR 2025 · similarity 0.85)
- OmniDrive: A Holistic Vision-Language Dataset for Autonomous Driving with Counterfactual Reasoning (CVPR 2025 · similarity 0.85)
- UniMotion: A Unified Motion Framework for Simulation, Prediction and Planning (NeurIPS 2025 · similarity 0.85)
- DriveSpatial: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving (NeurIPS 2026 · 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.
