What the Reranker Sees: Multi-Aspect Page Annotation for Long-Document Multimodal Question Answering

Annual reports, diligence packs, and infographic dashboards bury numbers in page images: axes, cell grids, and footnotes that OCR pipelines flatten and that page-level visual retrievers still treat as interchangeable in-context examples. We keep a frozen Qwen2.5-VL-7B-Instruct generator and a ColPali / VisRAG-Ret page index, and insert three modules. A capability-aware visual router (CAVR) tags each retrieved page as text, table, chart, layout, or mixed and mixes specialist experts before generation. Weak-to-strong page selection (WSPS) distils a frozen 7B answerability teacher into a 3B selection head so ranking is no longer a single InfoNCE score. Visual evidence threading (VET) builds layout-anchored paths of length at most three and lets the generator read the thread rather than a flat top-$k$ list. On gold-page DocVQA / ChartQA / InfographicVQA the 7B system reaches 96.3 / 90.1 / 85.4. Under the VisRAG top-3 protocol the mean generation accuracy is 62.74 versus 59.39 for the same backbone with concatenation. On MMLongBench-Doc retrieve-then-read, F1 moves from 19.2 to 22.6 and multi-page accuracy from 16.4 to 21.2. ViDoRe nDCG@5 after WSPS reranking is 83.6, with TAT-DQA financial reports at 70.4.

Vision-Language Models (VLMs) are increasingly used for long-document processing, where the inputs combine text with charts, tables, figures, and complex layouts. Deploying them means choosing how to feed the document to the model, which retriever to use when only a subset of pages is sent, and whether to run the model agentically or as a static pipeline. We study these choices on two long-document QA benchmarks with both frontier API and open-weight VLMs. First, on MMLongBench-Doc our six-tool agent with page, table, figure, and search calls pays off only once the answering VLM is large enough: with Qwen3.5-4B and 9B it trails static page input, with Qwen3.5-27B it draws level, and with Sonnet 4.5 it leads. On LongDocURL it is level with or ahead of static input at every reader. Its lead over the strongest static pipeline is clearest with the frontier reader on MMLongBench-Doc and narrows to within noise on LongDocURL. Second, retrieval modality matters more than the specific retriever: the strongest image retriever leads the strongest text pipeline, and on the text side a single off-the-shelf cross-encoder rerank essentially matches a much heavier multi-stage LLM pipeline. Top-k image retrieval is also the most token-efficient input at every reader we paired it with, at roughly a seventh to a quarter of the tokens of sending every page. Third, cutting across all three choices, three of our strongest pipelines succeed on different questions, and an oracle that picks the best pipeline per question gains roughly thirteen points over the best single pipeline, though evidence-type routing recovers almost none of it.

Real-world documents combine text with tables, charts, photographs, and diagrams arranged in diverse layouts, yet existing research on multimodal large language models (MLLMs) for document QA predominantly produces text-only responses, underutilizing these visual elements. We introduce VinQA, a dataset for long-form answer generation where cited visual elements are explicitly interleaved with their supporting text and grounded in relevant document pages. To support this task, we study two encoding methods for feeding raw document page images into an MLLM, along with their visual-element citation mechanisms: (1) Page Encoding, which directly encodes full-page images with bounding boxes of visual elements and treats these boxed regions as citable units; and (2) Modality Encoding, which parses each page to extract text and crop visual elements, encodes them separately, and uses these cropped elements as citable units. In our experiments, we propose M-GroSE, a multimodal evaluation framework extending GroUSE to assess answers along four dimensions: completeness, answer relevancy, faithfulness, and unanswerability. We additionally report Visual Source F1 to directly measure visual citation accuracy. Although proprietary frontier models still achieve the best overall scores on the VinQA test split, fine-tuning open Qwen2.5-VL models on the training split substantially improves their performance and narrows this gap. Modality Encoding is initially more robust for complex documents with long text, many visual elements, and diverse citation requirements. After training on VinQA, however, Page Encoding reaches a comparable level, competing effectively even without the explicit parsing used in Modality Encoding. Finally, Visual G-Eval, an MLLM-based judge, confirms that fine-tuned models insert visual elements at semantically appropriate positions with faithful supporting text.