Research map / Vision-language and multimodal / Multimodal LLMs
ROSE: Real-Time Open-World Scene Understanding from Monocular Video via Compact Multimodal 4D Scene Graphs
Ziyue Qiu, Yong Wang, Jin Pan
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 4DSG in prompt at inference (training-free) | DSR Qwen2.5-VL-7B · DSR-Bench (accuracy, 13 subtasks avg) | accuracy: 58.9 -> 62.3 | Improved |
| add 4DSG as context during fine-tuning | ROSE (Training-free) · DSR-Bench (accuracy, 13 subtasks avg) | accuracy: 62.3 -> 67.5 | Improved |
| add token-matched null context (LenCtrl) | DSR · DSR-Bench (accuracy, 13 subtasks avg) | accuracy: 58.9 -> 59.5 | No clear effect |
| add 4DSG fine-tuning | DSR · DSR-Bench held-out subtask family | accuracy: 22.9 -> 45.4 | Improved |
| add 4DSG fine-tuning | DSR · DSR-Bench Absolute Direction | accuracy: 73.8 -> 83.7 | Improved |
| add 4DSG fine-tuning | DSR · DSR-Bench Direction Prediction | accuracy: 35.5 -> 38.6 | Improved |
| add 2D anchors to raw video (ablation) | Qwen2.5-VL-32B · VLM4D balanced 200-question subset (Qwen2.5-VL-32B) | overall accuracy: 49.5 -> 56.0 | Improved |
| add 3D geometry and trajectories to anchors (ablation) | Qwen2.5-VL-32B · VLM4D balanced 200-question subset (Qwen2.5-VL-32B) | overall accuracy: 56.0 -> 64.5 | Improved |
| use bounding-box context crops instead of masked crops (ablation) | Qwen2.5-VL-32B 4DSG · VLM4D balanced 200-question subset (Qwen2.5-VL-32B) | overall accuracy: 64.5 -> 73.5 | Improved |
| use bounding-box context crops instead of masked crops (ablation) | Qwen2.5-VL-32B 4DSG · VLM4D FP | accuracy: 91.1 -> 91.1 | No clear effect |
| append 4DSG to prompt | Qwen2.5-VL-32B · DSR-Bench prompt latency | prompt latency (s): 3.95 -> 12.45 | Hurt |
| use ROSE (monocular 4DSG) instead of SNOW (RGB plus point clouds) | 4DSG pipeline · VLM4D Direct. | accuracy: 71.8 -> 79.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SNOW — SNOW: Spatio-Temporal Scene Understanding with World Knowledge for Open-World Embodied Reasoning (ECCV 2026) · better on 3 of 5 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
- GPT4Scene: Understand 3D Scenes from Videos with Vision-Language Models (ICLR 2026 · similarity 0.89)
- ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation (CVPR 2026 · similarity 0.89)
- MVVBench: Benchmarking 4D Reasoning in Vision-Language Models (NeurIPS 2026 · similarity 0.89)
- Motion4D: Learning 3D-Consistent Motion and Semantics for 4D Scene Understanding (NeurIPS 2025 · similarity 0.89)
- 4DP-QA: Scalable QA for 4D Perception in Vision Language Models (CVPR 2026 · similarity 0.88)
- InfiniBench: Infinite Benchmarking for Visual Spatial Reasoning with Customizable Scene Complexity (CVPR 2026 · 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.
