DataEvolver: Self-Evolving Multi-Agent Data Construction for Text-Rich Image Generation

• URL: http://arxiv.org/abs/2606.31537v1
• Date: Tue, 30 Jun 2026 11:52:22 GMT
• Title: DataEvolver: Self-Evolving Multi-Agent Data Construction for Text-Rich Image Generation
• Abstract summary: We propose DataEvolver, a framework for text-rich image data construction.<n>A Retriever collects candidate samples, a Verifier assigns quality scores and rejection causes, a Critic summarizes round-level feedback into semantic feedback, and a Generator completes under-covered regions through targeted synthesis.<n>Experiments show DataEvolver produces more useful training data than fixed-dataset baselines under matched data budgets.
• Score: 11.810930790760848
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Text-rich image generation is one of the most challenging settings in image generation, since models must simultaneously produce visually realistic images and render legible, semantically aligned, and layout-consistent text. Existing data pipelines usually follow a static crawl-filter-freeze paradigm. They collect candidate samples, filter them once, and freeze the accepted data for training. However, rejected samples are usually discarded, although they often contain useful failure signals such as OCR errors and semantic mismatches. As a result, later construction rounds may repeat the same failure modes. To address these limitations, we propose DataEvolver, a self-evolving multi-agent framework for text-rich image data construction. DataEvolver treats data construction as feedback-driven construction policy evolution. A Retriever collects candidate samples, a Verifier assigns quality scores and rejection causes, a Critic summarizes round-level feedback into semantic feedback, and a Generator completes under-covered regions through targeted synthesis. The updated feedback memory then guides the next construction round. Experiments on text-rich image generation benchmarks show that DataEvolver produces more useful training data than fixed-dataset baselines under matched data budgets. At the 0.75M scale on PixArt-alpha, DataEvolver improves OCR-F1 over the strongest baseline by 85.3 percent on TextScenesHQ and 35.3 percent on LongTextBench. The improvements are consistent across both evaluated benchmarks and also transfer to Show-o2, indicating that the benefit of DataEvolver is not tied to a single downstream generator. These results suggest that rejected samples can provide actionable feedback for improving text-rich image data construction.

Related papers

• Generative Retrieval for Unsupervised Text-Based Person Search [77.46429603553388]
  Text-based person search (TBPS) aims to retrieve images of a target person from a large image gallery based on a given natural language description.<n>We propose GTR+, a two-stage generation-then-retrieval framework.<n>We also contribute LargeFine-Person, a large-scale TBPS dataset with high-quality, fine-grained, and diverse textual annotations.
  arXiv  Detail & Related papers  (2026-09-11T15:25:00Z)
• VeriEvol: Scaling Multimodal Mathematical Reasoning via Verifiable Evol-Instruct [71.22358682581215]
  Scaling reinforcement learning for visual mathematical reasoning requires more than generating harder questions.<n>We instantiate this as VeriEvol, an iterative framework with two components.<n>On a five-benchmark visual-math suite, scaling evolved SFT data from 10K to 250K samples raises the mean accuracy from 35.42 to 54.73.
  arXiv  Detail & Related papers  (2026-06-22T16:17:30Z)
• TextSculptor: Training and Benchmarking Scene Text Editing [88.11688559021628]
  We present TextSculptor, a comprehensive framework for data construction and evaluation of scene text editing.<n>TextSculptor improves open-source text editing performance and narrows the gap to proprietary models.
  arXiv  Detail & Related papers  (2026-05-20T12:22:26Z)
• CycleCap: Improving VLMs Captioning Performance via Self-Supervised Cycle Consistency Fine-Tuning [65.10059440725041]
  Visual-Language Models (VLMs) have achieved remarkable progress in image captioning, visual question answering, and visual reasoning.<n>They remain prone to vision-language misalignment, often producing overly generic or hallucinated descriptions.<n>Existing approaches address this via instruction tuning-requiring costly, large-scale annotated datasets or via complex test-time frameworks for caption refinement.<n>In this work, we revisit image-text alignment through the lens of cycle consistency.
  arXiv  Detail & Related papers  (2026-03-18T20:57:31Z)
• When Vision Meets Texts in Listwise Reranking [1.2691047660244335]
  Rank-Nexus is a multimodal image-text document reranker that performs listwise qualitative reranking on retrieved lists incorporating both images and texts.<n>We first train modalities separately: leveraging abundant text reranking data, we distill knowledge into the text branch.<n>For images, where data is scarce, we construct distilled pairs from multimodal large language model (MLLM) captions on image retrieval benchmarks.
  arXiv  Detail & Related papers  (2026-01-28T13:57:14Z)
• Recycling the Web: A Method to Enhance Pre-training Data Quality and Quantity for Language Models [92.85086256871027]
  We propose REWIRE, REcycling the Web with guIded REwrite, to enrich low-quality documents so that they could become useful for training.<n>We demonstrate that about 82% of the mixed in texts come from transforming lower-quality documents that would otherwise be discarded.
  arXiv  Detail & Related papers  (2025-06-05T07:12:12Z)
• Cycle Consistency as Reward: Learning Image-Text Alignment without Human Preferences [23.022496443851235]
  Given an image and generated text, we map the text back to image space using a text-to-image model and compute the similarity between the original image and its reconstruction.<n>We use the cycle consistency score to rank candidates and construct a preference dataset of 866K comparison pairs.<n>The reward model trained on our dataset, CycleReward, outperforms state-of-the-art alignment metrics on detailed captioning.
  arXiv  Detail & Related papers  (2025-06-02T17:42:58Z)
• Data Extrapolation for Text-to-image Generation on Small Datasets [3.7356387436951146]
  We propose a new data augmentation method for text-to-image generation using linear extrapolation. We construct training samples dozens of times larger than the original dataset. Our model achieves FID scores of 7.91, 9.52 and 5.00 on the CUB, Oxford and COCO datasets.
  arXiv  Detail & Related papers  (2024-10-02T15:08:47Z)
• Re-Imagen: Retrieval-Augmented Text-to-Image Generator [58.60472701831404]
  Retrieval-Augmented Text-to-Image Generator (Re-Imagen) Retrieval-Augmented Text-to-Image Generator (Re-Imagen)
  arXiv  Detail & Related papers  (2022-09-29T00:57:28Z)

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.