Caption First, VQA Second: Knowledge Density, Not Task Format, Drives Multimodal Scaling

• URL: http://arxiv.org/abs/2604.13054v1
• Date: Tue, 17 Mar 2026 15:45:21 GMT
• Title: Caption First, VQA Second: Knowledge Density, Not Task Format, Drives Multimodal Scaling
• Abstract summary: We argue that the primary bottleneck in multimodal scaling is not task format, but knowledge density in training data.<n>We demonstrate that increasing knowledge density -- through structured caption enrichment and cross-modal knowledge injection -- leads to consistent performance improvements.
• Score: 5.3751181534542
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multimodal large language models (MLLMs) have achieved rapid progress, yet their scaling behavior remains less clearly characterized and often less predictable than that of text-only LLMs. Increasing model size and task diversity often yields diminishing returns. In this work, we argue that the primary bottleneck in multimodal scaling is not task format, but knowledge density in training data. We first show that task-specific supervision such as Visual Question Answering (VQA) contributes little incremental semantic information beyond image captions: VQA signals can be reconstructed from captions with negligible performance loss. We then demonstrate that increasing knowledge density -- through structured caption enrichment and cross-modal knowledge injection -- leads to consistent performance improvements across multimodal and downstream benchmarks. Across controlled experiments, performance correlates more strongly with semantic coverage than with task diversity. These findings suggest that current MLLMs fail to scale primarily because training data lacks sufficient knowledge coverage. We advocate for knowledge-centric multimodal training as a principled foundation for scalable multimodal models.

Related papers

• WeMMU: Enhanced Bridging of Vision-Language Models and Diffusion Models via Noisy Query Tokens [69.97021957331326]
  We propose Noisy Query Tokens, which learn a distributed representation space between the VLM and Diffusion Model via end-to-end optimization.<n>We also introduce a VAE branch with linear projection to recover fine-grained image details.
  arXiv  Detail & Related papers  (2025-12-02T09:02:20Z)
• A Comprehensive Study on Visual Token Redundancy for Discrete Diffusion-based Multimodal Large Language Models [85.30893355216486]
  We study how visual token redundancy evolves with different dMLLM architectures and tasks.<n>Our study reveals that visual redundancy emerges only in from-scratch dMLLMs while handling long-answer tasks.<n>Layer-skipping is promising for accelerating AR-to-diffusion dMLLMs, whereas progressive or late-step pruning is more effective for from-scratch dMLLMs.
  arXiv  Detail & Related papers  (2025-11-19T04:13:36Z)
• MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering [36.80441487363007]
  MLLMEraser is an input-aware, training-free framework for test-time unlearning.<n>We construct a multimodal erasure direction by contrasting adversarially perturbed, knowledge-recall image-text pairs.<n>Experiments on LLaVA-1.5 and Qwen-2.5-VL demonstrate that MLLMEraser consistently outperforms state-of-the-art MLLM unlearning baselines.
  arXiv  Detail & Related papers  (2025-10-05T14:20:17Z)
• True Multimodal In-Context Learning Needs Attention to the Visual Context [69.63677595066012]
  Multimodal Large Language Models (MLLMs) have enabled Multimodal In-Context Learning (MICL)-adapting to new tasks.<n>Current MLLMs tend to neglect visual cues and over-rely on textual patterns, leading to mere text imitation rather than genuine multimodal adaptation.<n>We introduce Dynamic Attention Reallocation (DARA), an efficient fine-tuning strategy that encourages models to attend to the visual context.
  arXiv  Detail & Related papers  (2025-07-21T17:08:18Z)
• MoCa: Modality-aware Continual Pre-training Makes Better Bidirectional Multimodal Embeddings [75.0617088717528]
  MoCa is a framework for transforming pre-trained VLM backbones into effective bidirectional embedding models.<n>MoCa consistently improves performance across MMEB and ViDoRe-v2 benchmarks, achieving new state-of-the-art results.
  arXiv  Detail & Related papers  (2025-06-29T06:41:00Z)
• DeepInsert: Early Layer Bypass for Efficient and Performant Multimodal Understanding [26.39397960987363]
  We propose a simple modification to pretrained transformer models.<n>Instead of concatenation with the language prompt at the start, we insert multimodal tokens directly into the middle.<n>Our results indicate that our method reduces computational costs during both training and inference.
  arXiv  Detail & Related papers  (2025-04-27T18:56:26Z)
• Learning to Instruct for Visual Instruction Tuning [79.76580302615174]
  We propose L2T, an advancement of visual instruction tuning (VIT)<n>L2T adopts a simple yet effective approach by incorporating the loss function into both the instruction and response sequences.<n>Surprisingly, L2T attains exceptional fundamental visual capabilities, yielding up to an 18% improvement in captioning performance.
  arXiv  Detail & Related papers  (2025-03-28T08:04:51Z)
• Visual RAG: Expanding MLLM visual knowledge without fine-tuning [5.341192792319891]
  This paper introduces Visual RAG, that synergically combines the MLLMs capability to learn from the context, with a retrieval mechanism.<n>In this way, the resulting system is not limited to the knowledge extracted from the training data, but can be updated rapidly and easily without fine-tuning.<n>It greatly reduces the computational costs for improving the model image classification performance, and augments the model knowledge to new visual domains and tasks it was not trained for.
  arXiv  Detail & Related papers  (2025-01-18T17:43:05Z)
• RA-BLIP: Multimodal Adaptive Retrieval-Augmented Bootstrapping Language-Image Pre-training [55.54020926284334]
  Multimodal Large Language Models (MLLMs) have recently received substantial interest, which shows their emerging potential as general-purpose models for various vision-language tasks. Retrieval augmentation techniques have proven to be effective plugins for both LLMs and MLLMs. In this study, we propose multimodal adaptive Retrieval-Augmented Bootstrapping Language-Image Pre-training (RA-BLIP), a novel retrieval-augmented framework for various MLLMs.
  arXiv  Detail & Related papers  (2024-10-18T03:45:19Z)
• MLLM Is a Strong Reranker: Advancing Multimodal Retrieval-augmented Generation via Knowledge-enhanced Reranking and Noise-injected Training [9.023648972811458]
  RagVL is a novel framework with knowledge-enhanced reranking and noise-injected training. We instruction-tune the MLLM with a simple yet effective instruction template to induce its ranking ability. For generation, we inject visual noise during training at the data and token levels to enhance the generator's robustness.
  arXiv  Detail & Related papers  (2024-07-31T08:43:17Z)
• Optimization of Prompt Learning via Multi-Knowledge Representation for Vision-Language Models [26.964848679914354]
  CoKnow is a framework that enhances Prompt Learning for Vision-Language Models with rich contextual knowledge.<n>We conducted extensive experiments on 11 publicly available datasets, demonstrating that CoKnow outperforms a series of previous methods.
  arXiv  Detail & Related papers  (2024-04-16T07:44:52Z)

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.