Beyond Static Visual Tokens: Structured Sequential Visual Chain-of-Thought Reasoning
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914427145551872 |
|---|---|
| author | Guo, Guangfu Lu, Xiaoqian Feng, Yue Sun, Mingming |
| author_facet | Guo, Guangfu Lu, Xiaoqian Feng, Yue Sun, Mingming |
| contents | Current multimodal LLMs encode images as static visual prefixes and rely on text-based reasoning, lacking goal-driven and adaptive visual access. Inspired by human visual perception-where attention is selectively and sequentially shifted from the most informative regions to secondary cues-we propose Structural Sequential Visual CoT SSV-CoT. First, a question-relevant saliency map identifies and organizes key visual regions, explicitly modeling the spatial distribution of visual importance. Second, reasoning is performed following this discriminative order, inducing a curriculum-like semantic progression from primary to secondary cues. This method is trained end-to-end, using text cot and answer supervision, without relying on region-level annotations or specialized external tools. Experiments on diverse visual reasoning benchmarks show gains, validating structured and sequential visual cognition. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_26737 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Beyond Static Visual Tokens: Structured Sequential Visual Chain-of-Thought Reasoning Guo, Guangfu Lu, Xiaoqian Feng, Yue Sun, Mingming Computer Vision and Pattern Recognition Artificial Intelligence Current multimodal LLMs encode images as static visual prefixes and rely on text-based reasoning, lacking goal-driven and adaptive visual access. Inspired by human visual perception-where attention is selectively and sequentially shifted from the most informative regions to secondary cues-we propose Structural Sequential Visual CoT SSV-CoT. First, a question-relevant saliency map identifies and organizes key visual regions, explicitly modeling the spatial distribution of visual importance. Second, reasoning is performed following this discriminative order, inducing a curriculum-like semantic progression from primary to secondary cues. This method is trained end-to-end, using text cot and answer supervision, without relying on region-level annotations or specialized external tools. Experiments on diverse visual reasoning benchmarks show gains, validating structured and sequential visual cognition. |
| title | Beyond Static Visual Tokens: Structured Sequential Visual Chain-of-Thought Reasoning |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2603.26737 |