[
    {
        "id": "osp-20169",
        "type": "article-journal",
        "title": "Rethinking Multi-Image Re-Representation in Multi-Image Understanding",
        "author": [
            {
                "family": "Zhang",
                "given": "Gengyuan"
            },
            {
                "family": "Han",
                "given": "Xiao"
            },
            {
                "family": "Xie",
                "given": "Xinyu"
            },
            {
                "family": "Liu",
                "given": "Tong"
            },
            {
                "family": "Tresp",
                "given": "Volker"
            }
        ],
        "URL": "https://omanscience.com/en/articles/rethinking-multi-image-re-representation-in-multi-image-understanding",
        "language": "en",
        "issued": {
            "date-parts": [
                [
                    2026
                ]
            ]
        },
        "abstract": "Multi-image understanding requires MLLMs not only to recognise the content of individual images, but also to organise visual evidence distributed across them. We study this problem through multi-image re-representation, viewing prompted Chain-of-Thought reasoning and agentic visual tool use as different ways of re-organising visual evidence during reasoning. We introduce Mosaic, a general-purpose multi-image visual harness that enables an MLLM to actively construct visual intermediates with ten composable image operations. We compare five re-representation settings on existing multi-image benchmarks and on MosaicBench, a new grounding-focused benchmark for fine-grained multi-image understanding. Our experiments show that the relative benefits of textual and visual re-representation are strongly task-dependent. Visual re-representation is particularly effective for tasks requiring precise visual evidence, including hypothesis testing, precision comparison, and orientation-sensitive reasoning, while tasks dominated by higher-level semantic content show smaller or less consistent gains. Building on this finding, we train MosaicAgent-8B to use Mosaic with reinforcement learning using only accuracy and format rewards. Without demonstration trajectories or rewards for specific tool-use, the agent learns to compose visual operations over multiple steps and exhibits diverse problem-solving patterns unpromptedly. Code and data will be released at https://github.com/gengyuanmax/Mosaic."
    }
]