arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2604.03476cs.CVcs.AIq-bio.BM

为分子结构识别微调DeepSeek-OCR-2

Fine-tuning DeepSeek-OCR-2 for Molecular Structure Recognition

Department of Computer Science, Princeton University, Princeton, NJ, USA（普林斯顿大学计算机科学系）；School of Pharmacy, University of Pittsburgh, Pittsburgh, PA, USA（匹兹堡大学药学院）；Khoury College of Computer Science, Northeastern University, Boston, MA, USA（东北大学Khoury计算机科学学院）；Computational Chemical Genomics Screening Center, University of Pittsburgh, Pittsburgh, PA, USA（匹兹堡大学计算化学基因组筛选中心）

Haocheng Tang, Xingyu Dang, Junmei Wang

阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出通过图像条件生成SMILES的方法微调DeepSeek-OCR-2，采用分阶段监督微调策略提升稳定性，训练数据结合合成图像和专利图像，模型在精确匹配上表现良好但不如图结构模型。

摘要：

光学化学结构识别（OCSR）对于将印刷文献中的2D分子图转化为机器可读格式至关重要。尽管视觉语言模型在端到端OCR任务中表现出色，但直接应用于OCSR仍具挑战性，全参数监督微调常失效。本文通过将任务转化为图像条件SMILES生成来适应DeepSeek-OCR-2进行分子光学识别。为克服训练不稳定性，我们提出双阶段渐进监督微调策略：先使用参数高效的LoRA，再通过分割学习率进行选择性全参数微调。我们训练模型于结合PubChem合成图像和USPTO-MOL现实专利图像的大型语料库以提升覆盖率和鲁棒性。我们的微调模型MolSeek-OCR展现出竞争力，精确匹配准确率可与最佳的图像到序列模型相媲美。然而，它仍劣于最先进的图像到图结构模型。此外，我们探索了强化学习风格的后训练和数据整理基于的改进，发现它们无法提高严格序列层面的忠实度以满足精确SMILES匹配需求。

英文摘要：

Optical Chemical Structure Recognition (OCSR) is critical for converting 2D molecular diagrams from printed literature into machine-readable formats. While Vision-Language Models have shown promise in end-to-end OCR tasks, their direct application to OCSR remains challenging, and direct full-parameter supervised fine-tuning often fails. In this work, we adapt DeepSeek-OCR-2 for molecular optical recognition by formulating the task as image-conditioned SMILES generation. To overcome training instabilities, we propose a two-stage progressive supervised fine-tuning strategy: starting with parameter-efficient LoRA and transitioning to selective full-parameter fine-tuning with split learning rates. We train our model on a large-scale corpus combining synthetic renderings from PubChem and realistic patent images from USPTO-MOL to improve coverage and robustness. Our fine-tuned model, MolSeek-OCR, demonstrates competitive capabilities, achieving exact matching accuracies comparable to the best-performing image-to-sequence model. However, it remains inferior to state-of-the-art image-to-graph modelS. Furthermore, we explore reinforcement-style post-training and data-curation-based refinement, finding that they fail to improve the strict sequence-level fidelity required for exact SMILES matching.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。

↑