Research map / Vision-language and multimodal / Multimodal LLMs
SpaceEra++: A Unified Framework Toward 3D Spatial Reasoning in Video
Weili Guan, H ZHANG, Meng Liu, Qianlong Xiang, Yue Wang, Liqiang Nie
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 ScenePick frame sampling | SpaceEra · VSI-Bench (Qwen2.5-VL-7B) | average accuracy: 43.9 -> 45.2 | Improved |
| add SpaceAlign RL training | SpaceEra · VSI-Bench (Qwen2.5-VL-7B) | average accuracy: 43.9 -> 48.7 | Improved |
| combine ScenePick and SpaceAlign | SpaceEra · VSI-Bench (Qwen2.5-VL-7B) | average accuracy: 43.9 -> 50.3 | Improved |
| add object semantics to ScenePick (not coverage only) (ablation) | ScenePick · VSI-Bench (Qwen2.5-VL-7B) | average accuracy: 44.9 -> 45.2 | Improved |
| add spatial coverage to ScenePick (not object semantics only) (ablation) | ScenePick · VSI-Bench (Qwen2.5-VL-7B) | average accuracy: 44.3 -> 45.2 | Improved |
| add space imagination rewards (R_abs and R_rel) (ablation) | SpaceAlign · VSI-Bench (Qwen2.5-VL-7B) | average accuracy: 45.8 -> 48.7 | Improved |
| add absolute coordinate reward (ablation) | SpaceAlign · VSI-Bench (Qwen2.5-VL-7B) | average accuracy: 46.3 -> 48.7 | Improved |
| add relative relation reward (ablation) | SpaceAlign · VSI-Bench (Qwen2.5-VL-7B) | average accuracy: 47.9 -> 48.7 | Improved |
| add ScenePick to SpaceEra on Gemini-1.5 Pro | SpaceEra · VSI-Bench (Gemini-1.5 Pro) | average accuracy: 52.8 -> 49.0 | Hurt |
| add ScenePick to SpaceEra on Qwen2.5-VL-72B | SpaceEra · VSI-Bench (Qwen2.5-VL-72B) | average accuracy: 47.1 -> 50.0 | Improved |
| add SpaceAlign to SpaceEra on Qwen3-VL-8B | SpaceEra · VSI-Bench (Qwen3-VL-8B) | average accuracy: 44.2 -> 49.2 | Improved |
| replace GPT-4o recognizer with Qwen3-VL-8B in SpatialMind (ablation) | SpatialMind · VSI-Bench (Qwen2.5-VL-72B) | average accuracy: 44.0 -> 42.3 | Hurt |
Closest papers by meaning
- SpaceVista: All-Scale Visual Spatial Reasoning from mm to km (ICML 2026 · similarity 0.93)
- Spatial Understanding from Videos: Structured Prompts Meet Simulation Data (NeurIPS 2025 · similarity 0.92)
- SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- RoboSpatial: Teaching Spatial Understanding to 2D and 3D Vision-Language Models for Robotics (CVPR 2025 · similarity 0.91)
- SpatiaLab: Can Vision–Language Models Perform Spatial Reasoning in the Wild? (ICLR 2026 · similarity 0.90)
- InternSpatial: A Comprehensive Dataset for Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.90)
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.
