arXiv Computer Vision October 28, 2026 By Zihu Wang, Karthik Somayaji N. S, Hejia Geng, Peng Li

ReGuLaR: Relation-Grounded Latent Reasoning for Large Vision-Language Models

Read the original on arXiv Computer Vision →

ReGuLaR introduces a relation‑grounded latent reasoning framework for large vision‑language models, explicitly anchoring latent states to key objects and their inter‑object relations. It employs a training‑time ReGFormer to focus reasoning on question‑relevant visual evidence, while inference proceeds without the ReGFormer. The authors also create the RGROUNDING‑351K dataset and demonstrate that ReGuLaR outperforms existing methods across diverse benchmarks, achieving state‑of‑the‑art results.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computer Vision.

arXiv Computation and Language

Sep 1

Decompose, Look, and Reason: Reinforced Latent Reasoning for VLMs

arXiv:2604.07518v2 Announce Type: replace Abstract: Vision-Language Models often struggle with complex visual reasoning due to the visual information loss in textual CoT. Existing methods either add...

By Mengdan Zhu, Senhao Cheng, Liang Zhao

llms rag diffusion multimodal benchmarks safety

arXiv Computation and Language

4d ago

Latent Visual Contrastive Reasoning with Multi-Branch Exploration

The paper introduces Latent Visual Contrastive Reasoning, a branch-aware latent reasoning framework that generates multiple question-conditioned latent visual branches as soft hypotheses over image tokens. It trains these branches using bounding‑box evidence annotations and a soft contrastive branch‑evidence alignment objective, encouraging evidence‑relevant branches to align with annotated visual regions while contrasting against negative evidence. Experiments on perception‑intensive visual reasoning benchmarks demonstrate that the method improves performance over the corresponding backbone and single‑path latent reasoning baselines at both 3B and 7B scales.

By Kegeng Tang, Jingbo Wang, Yangming Zhang, Huanshu Zhang, Shaogang Ren, Zihao Wang

llms diffusion nlp fine-tuning multimodal benchmarks safety

arXiv AI

Oct 9

SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows

SLVR (Structured Latent Visual Reasoning) is a training framework that blends explicit chain-of-thought with latent reasoning for multimodal large language models. It organizes reasoning into typed latent stages—planning, grounding, evidence selection, and integration—each supervised by corresponding signals such as plans, bounding boxes, visual evidence, and final rationales. Built on Qwen2.5‑VL‑7B, SLVR consistently improves performance on multimodal reasoning benchmarks, achieving significant gains on MMVP, BLINK Relation, V*, MathVista, and ChartQA.

By Albert Gao, Bing Xue, Andrea Zanette

llms multimodal benchmarks

arXiv Machine Learning

Jul 28

LanteRn: Latent Visual Structured Reasoning

arXiv:2603. 25629v2 Announce Type: replace-cross Abstract: While language reasoning models excel in many tasks, visual reasoning remains challenging for current large multimodal models (LMMs).

By Andr\'e G. Viveiros, Nuno Gon\c{c}alves, Matthias Lindemann, Andr\'e Martins

llms rag diffusion reinforcement-learning fine-tuning multimodal benchmarks

arXiv Computer Vision

Sep 30

Leveraging Latent Visual Reasoning in Silence

arXiv:2605.18641v2 Announce Type: replace Abstract: Latent visual reasoning involves visual evidence more directly in multimodal reasoning by inserting continuous latent tokens before textual generat...

By Dongyao Zhu, Zhen Wang, Xi Xiao, Han Jiang, Saeed Vahidian, Wei-Lun Chao, Tanya Berger-Wolf, Yu Su, Raju Vatsavai, Jianyang Gu

reinforcement-learning multimodal benchmarks

arXiv Machine Learning

Jun 2

DeepLatent: Think with Images via Parallel Latent Visual Reasoning

arXiv:2606. 00562v1 Announce Type: cross Abstract: The emerging paradigm of "thinking with images" embeds visual states into intermediate reasoning steps, defining a new frontier for Vision-Language Models.

By Dongchen Lu, Zhimo Li, Mao Shu, Huo Cao

llms rag reinforcement-learning robotics efficiency multimodal benchmarks