Eliciting Complex Spatial Reasoning in MLLMs through Wide-Baseline Matching
About
Wide-baseline matching (WBM) requires integrating geometric understanding, viewpoint changes, fine-grained perception, and occlusion reasoning, making it a challenging testbed for spatial reasoning in multimodal large language models (MLLMs) deployed in physical environments. However, current MLLMs lack systematic evaluation and training frameworks for these capabilities. We introduce ReasonMatch-Bench, a benchmark stratified by viewpoint displacement and matching granularity across indoor, outdoor, and object-centric scenarios, and show that current MLLMs still struggle with fine-grained wide-baseline correspondence: on a difficult 90-sample subset, human annotators achieve 84.0 F1, while the best existing baseline reaches 37.2. To bridge this gap, we build a scalable data-generation pipeline that automatically extracts wide-baseline view pairs from large-scale video-3D corpora, including RGB-D videos and SfM reconstructions, yielding diverse and verifiable supervision. We further propose Dynamic Correspondence Reinforcement Learning (DCRL), which combines Image-Level Viewpoint Progression and Point-Level Correspondence Curriculum to improve WBM training through verifiable rewards without explicit CoT supervision. Extensive experiments show that DCRL substantially improves ReasonMatch-Bench and transfers to related spatial benchmarks, while maintaining general visual understanding performance with modest gains on several benchmarks.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| General Visual Understanding | RealworldQA | Accuracy70.5 | 64 | |
| Image Understanding | MMStar | Score62.5 | 56 | |
| Spatial Reasoning | SAT (real) | Accuracy (Pass@1)75.3 | 25 | |
| Wide-Baseline Correspondence Reasoning | ReasonMatch-Bench Overall 90-sample high-divergence subset | F1 Score70.6 | 18 | |
| Wide Baseline Matching | ReasonMatch-Bench Overall (test) | F1 Score70.5 | 10 | |
| Wide Baseline Matching | ReasonMatch-Bench Outdoor (test) | L1 Score90.9 | 10 | |
| Wide Baseline Matching | ReasonMatch-Bench Object (test) | L1 Metric45.6 | 10 | |
| Wide Baseline Matching | ReasonMatch-Bench Indoor (test) | L1 Error84.6 | 10 | |
| Spatial Reasoning | MindCube | Overall Accuracy43.52 | 9 | |
| Cross-view Correspondence | DL3DV ReasonMatch-Bench high-divergence samples | Precision53.3 | 6 |