LaRe: Latent Refocusing for Multimodal Reasoning
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866914603697438720 |
|---|---|
| author | Ma, Jizheng Zhou, Xiaofei Zhang, Geyuan Song, Yanlong Yan, Han |
| author_facet | Ma, Jizheng Zhou, Xiaofei Zhang, Geyuan Song, Yanlong Yan, Han |
| contents | Chain of Thought (CoT) reasoning enhances logical performance by decomposing complex tasks, yet its multimodal extension faces a trade-off. The prevailing Thinking with Images paradigm achieves visual refocusing by explicitly cropping image regions, yet incurs rapidly growing computational overhead. The emerging line of latent-space reasoning reduces token consumption, but lacks the capacity for dynamic refocusing. We argue that this trade-off stems from a tacitly accepted premise that effective visual refocusing must occur in the form of explicit tokens. Building on this, we propose Latent Refocusing (LaRe), a new multimodal reasoning paradigm in which visual refocusing takes place entirely within the latent space. We further design a semantic augmentation training strategy that ensures the semantic structure of the latent space through visual reconstruction objective. Experimental evaluations demonstrate that LaRe improves average accuracy by 7.6% compared to existing baselines while reducing the number of tokens required for inference by 59.7%. When scaled to a 8B-parameter Vision-Language Model backbone, LaRe achieves performance comparable to state-of-the-art methods, demonstrating the efficacy of our proposed latent refocusing paradigm for multimodal reasoning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_02360 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | LaRe: Latent Refocusing for Multimodal Reasoning Ma, Jizheng Zhou, Xiaofei Zhang, Geyuan Song, Yanlong Yan, Han Computer Vision and Pattern Recognition Computation and Language Chain of Thought (CoT) reasoning enhances logical performance by decomposing complex tasks, yet its multimodal extension faces a trade-off. The prevailing Thinking with Images paradigm achieves visual refocusing by explicitly cropping image regions, yet incurs rapidly growing computational overhead. The emerging line of latent-space reasoning reduces token consumption, but lacks the capacity for dynamic refocusing. We argue that this trade-off stems from a tacitly accepted premise that effective visual refocusing must occur in the form of explicit tokens. Building on this, we propose Latent Refocusing (LaRe), a new multimodal reasoning paradigm in which visual refocusing takes place entirely within the latent space. We further design a semantic augmentation training strategy that ensures the semantic structure of the latent space through visual reconstruction objective. Experimental evaluations demonstrate that LaRe improves average accuracy by 7.6% compared to existing baselines while reducing the number of tokens required for inference by 59.7%. When scaled to a 8B-parameter Vision-Language Model backbone, LaRe achieves performance comparable to state-of-the-art methods, demonstrating the efficacy of our proposed latent refocusing paradigm for multimodal reasoning. |
| title | LaRe: Latent Refocusing for Multimodal Reasoning |
| topic | Computer Vision and Pattern Recognition Computation and Language |
| url | https://arxiv.org/abs/2511.02360 |