Research map / Vision-language and multimodal / Multimodal LLMs
MV-STRIDE: Enabling MLLMs to Master Multi-View Spatial Reasoning via Hierarchical Capability Modeling
Jin Xu, Xiaojian Huang, zhang zhihong, Luo Zhuodong, Xuejin Chen, Jie Zhao, Xin Liu, Wang Xinzhi, Jiansheng Wei
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 Stage 1 foundational SFT before cold start and RL (ablation) | MV-STRIDE-RL · MMSI-Bench | accuracy: 29.20 -> 37.50 | Improved |
| add Stage 2 CoT cold start before RL (ablation) | MV-STRIDE-RL · CV-Bench | accuracy: 69.37 -> 85.94 | Improved |
| add Stage 2 CoT cold start before RL (ablation) | MV-STRIDE-RL · 3DSRBench | accuracy: 43.16 -> 59.43 | Improved |
| add Stage 3 RL after cold start (ablation) | MV-STRIDE · MMSI-Bench | accuracy: 36.70 -> 37.50 | Improved |
| add Stage 3 RL after cold start (ablation) | MV-STRIDE · 3DSRBench | accuracy: 61.08 -> 59.43 | Hurt |
| switch from short-answer SFT (Stage 1) to CoT cold start (Stage 2) (ablation) | MV-STRIDE · MMSI-Bench | accuracy: 37.20 -> 36.70 | Hurt |
| use all-in-one SFT instead of multi-stage training with RL | MV-STRIDE · MMSI-Bench | accuracy: 38.90 -> 37.50 | Hurt |
| combine synthetic and real-world Level III data instead of synthetic only (ablation) | MV-STRIDE SFT · MMSI-Bench | accuracy: 36.70 -> 39.40 | Improved |
| use real-world instead of synthetic data only (ablation) | MV-STRIDE SFT · MMSI-Bench | accuracy: 36.70 -> 37.50 | Improved |
| add Level I single-view perception data (ablation) | MV-STRIDE Stage 1 · ViewSpatial-Bench | accuracy: 48.77 -> 51.17 | Improved |
| add Level II data on top of Level III only (ablation) | MV-STRIDE Stage 1 · MMSI-Bench | accuracy: 36.80 -> 37.30 | Improved |
| fine-tune base model with MV-STRIDE data (SFT Full) | Qwen3-VL-8B · MMSI-Bench | accuracy: 29.20 -> 38.90 | Improved |
Closest papers by meaning
- Seeing from Another Perspective: Evaluating Multi-View Understanding in MLLMs (AAAI 2026 · similarity 0.92)
- InternSpatial: A Comprehensive Dataset for Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.91)
- STAR-R1: Multi-View Spatial TrAnsformation Reasoning by Reinforcing Multimodal LLMs (CVPR 2026 · similarity 0.91)
- Eliciting Complex Spatial Reasoning in MLLMs through Wide-Baseline Matching (CVPR 2026 · similarity 0.91)
- Spatial-MLLM: Boosting MLLM Capabilities in Visual-based Spatial Intelligence (NeurIPS 2025 · similarity 0.91)
- Multi-SpatialMLLM: Multi-Frame Spatial Understanding with Multi-Modal Large Language Models (CVPR 2026 · similarity 0.91)
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.
