See First, Answer Later: Visual Evidence Pre-Alignment via Sufficiency-Driven RL

• URL: http://arxiv.org/abs/2606.17678v1
• Date: Tue, 16 Jun 2026 08:45:24 GMT
• Title: See First, Answer Later: Visual Evidence Pre-Alignment via Sufficiency-Driven RL
• Abstract summary: Multimodal large language models (MLLMs) integrate strong text reasoning with visual inputs, yet their responses can be inconsistent with the underlying images.<n>We introduce Visual Evidence Pre-Alignment (VEPA), an intermediate stage between pretraining and post-training to optimize question-conditioned visual evidence descriptions.
• Score: 19.682815357013453
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multimodal large language models (MLLMs) integrate strong text reasoning with visual inputs, yet their responses can be inconsistent with the underlying images, indicating ineffective utilization of visual evidence during inference. The prevailing training paradigm relies on large-scale caption-based pretraining for general alignment, followed by supervised fine-tuning and reinforcement learning to enable instruction following and complex reasoning. However, such pretraining provides only weak visual grounding: short, coarse captions bias models toward salient objects while neglecting fine-grained visual evidence. In this paper, we introduce Visual Evidence Pre-Alignment (VEPA), an intermediate stage between pretraining and post-training that explores a novel sufficiency-driven objective with Group Relative Policy Optimization (GRPO) to optimize question-conditioned visual evidence descriptions. Extensive experiments across diverse benchmarks show that our VEPA consistently enhances performance on visually demanding evaluations and complements standard supervised post-training. Further analyses show that the income stems from strengthened, transferable visual grounding, rather than from additional task-specific training.

Related papers

• Be Faithful When Response: Returning Fluent and Grounded Answers for Vision-Language Models Reinforcement Learning [12.622982203412478]
  Reinforcement Learning (RL) is an important paradigm for improving the reasoning capabilities of Vision-Language Models (VLMs)<n>We propose a Faithful Warm-Start (FWS) strategy that first curates samples with explicit vision-language causal relationships from six general VQA benchmarks.<n>A VLM-based judge is employed to further purify the dataset, ensuring strong causal consistency and visual faithfulness.<n> Experimental results show that such faithful supervision improves answer accuracy, stabilizes RL training, and reduces visually unsupported reasoning.
  arXiv  Detail & Related papers  (2026-06-29T08:58:33Z)
• From Seeing to Thinking: Decoupling Perception and Reasoning Improves Post-Training of Vision-Language Models [66.95781712577315]
  We study the interplay between perception and reasoning in vision-language models (VLMs) post-training by decomposing their capabilities into three separate training stages.<n>We show that models trained with our approach achieve 1.5% higher reasoning accuracy with 20.8% shorter reasoning traces.
  arXiv  Detail & Related papers  (2026-05-19T17:58:40Z)
• Enhancing Multimodal In-Context Learning via Inductive-Deductive Reasoning [65.15766304205657]
  In-context learning (ICL) allows large models to adapt to tasks using a few examples, yet its extension to vision-language models (VLMs) remains fragile.<n>Our analysis reveals that the fundamental limitation lies in an inductive gap, models often produce correct answers from flawed reasoning.<n>We introduce a framework that restructures multimodal ICL as a principled inductive-deductive process.
  arXiv  Detail & Related papers  (2026-05-04T09:18:19Z)
• Visually-Guided Policy Optimization for Multimodal Reasoning [60.035908460318126]
  We propose Visually-Guided Policy Optimization (VGPO) to reinforce visual focus during policy optimization.<n> VGPO initially introduces a Visual Attention Compensation mechanism that leverages visual similarity to localize and amplify visual cues.<n>We show that VGPO achieves better visual activation and superior performance in mathematical multimodal reasoning and visual-dependent tasks.
  arXiv  Detail & Related papers  (2026-04-10T14:22:38Z)
• Selective Training for Large Vision Language Models via Visual Information Gain [7.834991119179473]
  We introduce Visual Information Gain (VIG), a perplexity-based metric.<n>VIG measures the reduction in prediction uncertainty provided by visual input.<n>We propose a VIG-guided selective training scheme that prioritizes high-VIG samples and tokens.
  arXiv  Detail & Related papers  (2026-02-19T09:12:21Z)
• MMRPT: MultiModal Reinforcement Pre-Training via Masked Vision-Dependent Reasoning [20.14427952871989]
  We introduce MMRPT, a multimodal reinforcement pre-training framework that strengthens visual reasoning in MLLMs.<n>We are the first to incorporate reinforcement learning directly into the pre-training of large vision-language models.<n>Experiments show consistent zero-shot gains across diverse benchmarks and substantially improved robustness under supervised fine-tuning.
  arXiv  Detail & Related papers  (2025-12-08T06:26:13Z)
• Rethinking Visual Intelligence: Insights from Video Pretraining [75.32388528274224]
  Large language models (LLMs) have demonstrated that large-scale pretraining enables systems to adapt rapidly to new problems.<n>We investigate Video Diffusion Models (VDMs) as a promising direction for bridging the gap.
  arXiv  Detail & Related papers  (2025-10-28T14:12:11Z)
• Learning to See Before Seeing: Demystifying LLM Visual Priors from Language Pre-training [37.93241751782069]
  Large Language Models (LLMs) develop rich visual priors despite being trained on text alone.<n>These priors allow latent visual capabilities to be unlocked for vision tasks with a relatively small amount of multimodal data.<n>We show that visual priors are composed of separable perception and reasoning priors with unique scaling trends and origins.
  arXiv  Detail & Related papers  (2025-09-30T17:57:44Z)
• Preserving Clusters in Prompt Learning for Unsupervised Domain Adaptation [29.809079908218607]
  This work introduces a fresh solution to reinforce base pseudo-labels and facilitate target-prompt learning.<n>We first propose to leverage the reference predictions based on the relationship between source and target visual embeddings.<n>We later show that there is a strong clustering behavior observed between visual and text embeddings in pre-trained multi-modal models.
  arXiv  Detail & Related papers  (2025-06-13T06:33:27Z)
• Debiasing Multimodal Large Language Models via Penalization of Language Priors [38.97645845493758]
  Multimodal Large Language Models (MLLMs) have become indispensable tools in computer vision and natural language processing.<n>Despite their advancements, our investigation reveals a noteworthy bias: the generated content is often driven more by the inherent priors of the underlying Large Language Models (LLMs) than by the input image.<n>We propose two simple, training-free strategies to rectify these biases and redirect the model's focus toward visual information.
  arXiv  Detail & Related papers  (2024-03-08T12:35:07Z)
• Weakly-Supervised Learning of Visual Relations in Multimodal Pretraining [25.11384964373604]
  We propose two pretraining approaches to contextualise visual entities in a multimodal setup. With verbalised scene graphs, we transform visual relation triplets into structured captions, and treat them as additional image descriptions. With masked relation prediction, we further encourage relating entities from image regions with visually masked contexts.
  arXiv  Detail & Related papers  (2023-05-23T17:27:12Z)
• SgVA-CLIP: Semantic-guided Visual Adapting of Vision-Language Models for Few-shot Image Classification [84.05253637260743]
  We propose a new framework, named Semantic-guided Visual Adapting (SgVA), to extend vision-language pre-trained models. SgVA produces discriminative task-specific visual features by comprehensively using a vision-specific contrastive loss, a cross-modal contrastive loss, and an implicit knowledge distillation. State-of-the-art results on 13 datasets demonstrate that the adapted visual features can well complement the cross-modal features to improve few-shot image classification.
  arXiv  Detail & Related papers  (2022-11-28T14:58:15Z)
• Dense Contrastive Visual-Linguistic Pretraining [53.61233531733243]
  Several multimodal representation learning approaches have been proposed that jointly represent image and text. These approaches achieve superior performance by capturing high-level semantic information from large-scale multimodal pretraining. We propose unbiased Dense Contrastive Visual-Linguistic Pretraining to replace the region regression and classification with cross-modality region contrastive learning.
  arXiv  Detail & Related papers  (2021-09-24T07:20:13Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.