Real-world vision-language applications demand varying levels of perceptual granularity. However, most existing visual large language models (VLLMs), such as LLaVA, pre-assume a fixed resolution for downstream tasks, which leads to subpar performance. To address this problem, we first conduct a comprehensive and pioneering investigation into the resolution preferences of different vision-language tasks, revealing a correlation between resolution preferences with (1) image complexity, and (2) uncertainty variance of the VLLM at different image input resolutions. Building on this insight, we propose an empirical formula to determine the optimal resolution for a given vision-language task, accounting for these two factors as the zeroth-order and first-order terms in the Taylor expansion on a given image input. Second, based on rigorous experiments, we propose a novel parameter-efficient fine-tuning technique to extend the visual input resolution of pre-trained VLLMs to the identified optimal resolution. Extensive experiments on various vision-language tasks validate the effectiveness of our method.

Multimodal Large Language Models (MLLMs) have gained increasing popularity as a promising framework for leveraging the strong language reasoning capabilities in the vision-language domain. Given a wide range of MLLMs, model merging potentially offers a cheap way to aggregate their diverse knowledge into a single MLLM. However, directly plug-in existing model merging approaches often leads to suboptimal performance due to (1) inclusion of harmful models that have over-confident predictions in the target task; (2) the lack of specialized designs for vision-language inputs. To tackle these pain points, we conduct pioneering investigations to dissect the merging procedures and propose an uncertainty-guided MLLM merging algorithm, i.e., UQ-Merge, which i) identifies beneficial candidates for merging, ii) determines the merging order and the number of helpful candidates, and iii) performs appropriate merging. Within our framework, we consider uncertainty quantification on both text and vision inputs to examine the MLLM prediction confidence, and then decide whether and when a MLLM needs to be included. It is worth mentioning that our vision-language uncertainty quantification does not require access to sample labels, making it more practical in various scenarios. Extensive experiments consistently demonstrate the superior MLLM merging performance of UQ-Merge in both held-in and held-out vision-language benchmarks. For example, compared to existing state-of-the-art merging methods, UQ-Merge brings substantial performance improvements of up to 44.3% on average accuracy in 12 datasets. Codes are available at https://anonymous.4open.science/r/UQ-Merge-7CD7.

Large Vision-Language Models often generate hallucinated content that is not grounded in its visual inputs. While prior work focuses on mitigating hallucinations, we instead explore leveraging hallucination correction as a training objective to improve video-language alignment. We introduce HACA, a self-training framework learning to correct hallucinations in descriptions that do not align with the video content. By identifying and correcting inconsistencies, HACA enhances the model’s ability to align video and textual representations for spatio-temporal reasoning. Our experimental results show consistent gains in video-caption binding and text-to-video retrieval tasks, demonstrating that hallucination correction-inspired tasks serve as an effective strategy for improving vision and language alignment.

In our work, we explore the synergistic capabilities of pre-trained vision-and-language models (VLMs) and large language models (LLMs) on visual commonsense reasoning (VCR) problems. We find that VLMs and LLMs-based decision pipelines are good at different kinds of VCR problems. Pre-trained VLMs exhibit strong performance for problems involving understanding the literal visual content, which we noted as visual commonsense understanding (VCU). For problems where the goal is to infer conclusions beyond image content, which we noted as visual commonsense inference (VCI), VLMs face difficulties, while LLMs, given sufficient visual evidence, can use commonsense to infer the answer well. We empirically validate this by letting LLMs classify VCR problems into these two categories and show the significant difference between VLM and LLM with image caption decision pipelines on two subproblems. Moreover, we identify a challenge with VLMs’ passive perception, which may miss crucial context information, leading to incorrect reasoning by LLMs. Based on these, we suggest a collaborative approach, named ViCor, where pre-trained LLMs serve as problem classifiers to analyze the problem category, then either use VLMs to answer the question directly or actively instruct VLMs to concentrate on and gather relevant visual elements to support potential commonsense inferences. We evaluate our framework on two VCR benchmark datasets and outperform all other methods without in-domain fine-tuning.