Research map / Vision-language and multimodal / Multimodal LLMs
StreamSpatial: A Benchmark and Framework for Streaming 3D Visual-Spatial Reasoning
Junlin Xie, Keyang Zhong, Quanlong Zheng, Ruifei Zhang, Kuo Wang, Yanhao Zhang, Haonan Lu, Xiang Wan, Guanbin Li
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 |
|---|---|---|---|
| apply Stream3D-Mem memory and latent cognitive-map reasoning | Qwen2.5-VL-3B · HourVideo Navigation | accuracy: 32.14 -> 37.50 | Improved |
| apply Stream3D-Mem | Qwen2.5-VL-3B · HourVideo Perception | accuracy: 35.10 -> 41.39 | Improved |
| apply Stream3D-Mem | Qwen2.5-VL-3B · HourVideo Reasoning | accuracy: 31.60 -> 35.32 | Improved |
| apply Stream3D-Mem | Qwen2.5-VL-3B · HourVideo Summarization | accuracy: 50.75 -> 47.56 | Hurt |
| add spatial-specific criteria to a simple prompt (ablation) | MLLMs · StreamSpatialBench | accuracy: <=4% | No clear effect |
| add Zero-CoT, streaming checkpoint CoT or memory-replay CoT prompting (ablation) | MLLMs · StreamSpatialBench | accuracy | No clear effect |
| increase temporal span and region span of the question (ablation) | MLLMs · StreamSpatialBench historical tasks | accuracy | Hurt |
| increase model scale within a family (ablation) | Qwen2.5-VL · StreamSpatialBench | avg accuracy: 34.49 -> 38.41 | Improved |
| increase model scale within a family (ablation) | LLaVA-Video · StreamSpatialBench | avg accuracy: 38.41 -> 44.01 | Improved |
| specialize for spatial (Spatial-MLLM) instead of general-purpose model (ablation) | 7B-8B models · StreamSpatialBench | avg accuracy: 33.28 -> 37.01 | Hurt |
| apply Stream3D-Mem to evaluate offline | Stream3D-Mem · VSI-Bench | average accuracy: 41.10 -> 47.30 | 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.
- Spatial-MLLM — Spatial-MLLM: Boosting MLLM Capabilities in Visual-based Spatial Intelligence (NeurIPS 2025) · better on 1 of 1 shared result
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
- MindJourney: Test-Time Scaling with World Models for Spatial Reasoning (NeurIPS 2025 · similarity 0.90)
- SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning (NeurIPS 2026 · similarity 0.90)
- SpatialReasoner: Towards Explicit and Generalizable 3D Spatial Reasoning (NeurIPS 2025 · similarity 0.89)
- InfiniBench: Infinite Benchmarking for Visual Spatial Reasoning with Customizable Scene Complexity (CVPR 2026 · similarity 0.89)
- ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation (CVPR 2026 · similarity 0.89)
- Spatial Understanding from Videos: Structured Prompts Meet Simulation Data (NeurIPS 2025 · similarity 0.89)
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.
