Improving chemical image recognition reliability with verification and revision
ViCoR: Reliable Molecular Structure Extraction via Spatially Aligned Verification and Executable Revision
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Chemical structures from scientific papers are often converted into computer-readable formats by machines, but mistakes in this process can cause problems in chemical databases. The authors created ViCoR, a system that checks and fixes errors in these chemical images by comparing the original picture with its computer interpretation. This method improves accuracy significantly and helps produce reliable chemical data without needing as much manual correction. Their approach also boosts the performance of other chemistry-related computer models.
What this means in practice
- •For chemical data engineers: Automatically verify and repair molecular images from literature to build high-quality chemical databases with less human effort.
- •For pharmaceutical ai developers: Improve chemical reaction extraction and prediction models by integrating more accurate molecular structure data from automated recognition systems.
Authors
Yujian Yuan, Xin Cai, Yufan Chen, Jiaxin Xu, Mengdi Liu, Zhichao Tan, Long Chen, Hanyu Gao
Abstract
Reliable optical chemical structure recognition (OCSR) is essential for building high-quality chemical data from scientific literature, yet even small recognition errors can propagate into chemical databases and downstream models. In practice, recognized structures often require manual inspection and correction before use, making large-scale data curation costly and difficult to scale. We therefore study Selective Structure Recognition (SSR), a post-recognition setting that automatically produces reliable structured outputs while rejecting unresolved cases. Selection-only approaches can improve reliability by rejection, but cannot create additional correct outputs beyond those produced by the base recognizer. We propose ViCoR, a repair-before-rejection framework for iterative VerIfiCatiOn and Revision. Its key idea is to make observation-prediction correspondence explicit: coordinate-preserving rendering establishes spatial correspondence between the source image and predicted structure, while index anchoring maps localized visual discrepancies to executable graph edits without full-structure regeneration. A shared VLM is progressively trained from verification to revision. On two real-world OCSR benchmarks, ViCoR improves overall accuracy from 73.53\% to 88.26\% and from 61.83\% to 84.32\%, while achieving over 97\% accepted accuracy at 85--89\% coverage. The resulting molecular data further improve reaction-extraction F1 by 15.5 points and literature-sourced reaction prediction accuracy by 7.7 and 5.8 points, demonstrating the value of automated reliability control for scientific data curation and downstream chemical learning.