MolParser-Mobile: Ultrafast OCSR System for Large-Scale Chemical Literature Mining

Xi Fang; Haocheng Lu; Han Lyu; Chengxiang Luo; Linfeng Zhang; Guolin Ke

cs.CVSep 5, 2026

MolParser-Mobile: Ultrafast OCSR System for Large-Scale Chemical Literature Mining

Authors: Xi Fang, Haocheng Lu, Han Lyu, Chengxiang Luo, Linfeng Zhang, Guolin Ke

Organizations: DP Technology

Abstract

Optical Chemical Structure Recognition (OCSR) is a fundamental component of chemical literature mining, enabling molecular database construction, reaction extraction, and AI-driven scientific discovery. Despite substantial progress in recognition accuracy with recent deep learning-based methods, inference throughput remains a critical bottleneck that limits web-scale deployment. To address this challenge, we propose MolParser-Mobile, an AutoML-optimized lightweight end-to-end OCSR framework. MolParser-Mobile contains only 9.98M parameters, while reaching a throughput of 1,520 molecules per second on a single NVIDIA RTX 4090D GPU. Despite its compact design, it maintains competitive and, on several benchmarks, superior recognition accuracy.

Figures & tables

Model	Params	Throughput	UniParserBench	BioVista
Gemma4-31B	31B	OOM	0.073	0.118
Qwen3.5-397B	397B	OOM	0.272	0.347
MolParser	216M	39.8	0.800	0.703
MolParser-Mobile	$\downarrow$ 95.4% 9.98M $\downarrow$ 95.4%	$\uparrow$ 38 $\times$ 1,520 $\uparrow$ 38 $\times$	$\uparrow$ 0.023 0.823 $\uparrow$ 0.023	$\uparrow$ 0.098 0.801 $\uparrow$ 0.098

Table 1

Figure 1: Accuracy–latency trade-off of molecular structure detectors. (a) MolDet-Bench-General evaluates molecular localization over diverse image types and scales, whereas (b) MolDet-Bench-Doc and (c) BioVista evaluate document-page inputs. MolDetv2 provides favorable Pareto operating points, achieving high

\mathrm{mAP}_{50}

with low inference latency. Legacy MolDet variants are connected according to model scale (

n\!\rightarrow\!s\!\rightarrow\!m\!\rightarrow\!l

), while MolDetv2 and Uni-Parser-LD are shown separately. Horizontal error bars indicate latency variation, and dashed lines in (c) denote methods for which latency is unavailable.

Model	$\mathrm{mAP}_{50}\uparrow$	$\mathrm{mAP}_{50:95}\uparrow$	Latency (ms) $\downarrow$
MolDet-Bench-General
MolDetv2-General-n	0.9872	0.8776	$\mathbf{1.5\pm 0.0}$
MolDet-General-m	0.9702	0.8269	$4.7\pm 0.1$
MolDet-General-l	0.9675	0.8349	$6.2\pm 0.1$
MolDet-General-n	0.9574	0.8052	$\mathbf{1.5\pm 0.0}$
MolDet-Bench-Doc

Table 1: Representative results on MolDet-Bench. We report latency-matched and strongest representative baselines; complete accuracy–latency comparisons are provided in Fig. 1 . The best accuracy and lowest latency in each benchmark are shown in bold. Latency is measured in milliseconds on an NVIDIA T4 using TensorRT 10.

Figure 2: MolGallery construction pipeline. Molecular images extracted from real-world chemical documents are recognized by three OCSR models, with samples retained when all three models produce identical E-SMILES predictions.

Platform	Hardware	Throughput (samples/s) $\uparrow$	Backend
GPU	NVIDIA RTX 4090D 24GB	1520	PyTorch
CPU	Intel Xeon Platinum 8336C, 14 vCPU, 2.30GHz	78.11	PyTorch

Table 2: Inference throughput of MolParser-Mobile. Throughput is measured in samples per second using the native PyTorch backend (no compile, FP16 in GPU and FP32 in CPU).

Optical Character Recognition

Optical character recognition converts text in images into machine-readable text, from printed pages and handwriting to signs and symbols.

See all 201 papers

COMO: Closed-Loop Optical Molecule Recognition with Minimum Risk Training
Apr 26, 2026Zhuoqi Lyu, Qing KeExposure Bias Optical Character Recognition
Department of Data Science, College of Computing, City University of Hong Kong, Hong Kong, China
MolRecBench-Wild: A Real-World Benchmark for Optical Chemical Structure Recognition
May 7, 2026Haote Yang, Hui Wang, Chen Zhu +14Optical Character Recognition
Shanghai Artificial Intelligence Laboratory · King’s College London · East China University of Science and Technology +5
Fine-tuning DeepSeek-OCR-2 for Molecular Structure Recognition
Apr 3, 2026Haocheng Tang, Xingyu Dang, Junmei WangOptical Character Recognition
School of Pharmacy, University of Pittsburgh, Pittsburgh, PA, USA · Computational Chemical Genomics Screening Center, University of Pittsburgh, Pittsburgh, PA, USA · Khoury College of Computer Science, Northeastern University, Boston, MA, USA +1

Apr 26, 2026J/K move · Enter open · S save

COMO: Closed-Loop Optical Molecule Recognition with Minimum Risk Training

Zhuoqi Lyu, Qing Ke

Department of Data Science, College of Computing, City University of Hong Kong, Hong Kong, China

Exposure Bias Optical Character Recognition

Open paper AskPDF arXiv

Optical chemical structure recognition (OCSR) translates molecular images into machine-readable representations like SMILES strings or molecular graphs, but remains challenging in real-world documents due to inexhaustible variations in chemical structures, shorthand conventions, and visual noise. Most existing deep-learning-based approaches rely on teacher forcing with token-level Maximum Likelihood Estimation (MLE). This training paradigm suffers from exposure bias, as models are trained under ground-truth prefixes but must condition on their own previous predictions during inference. Moreover, token-level MLE objectives hinder the optimization towards molecular-level evaluation criteria such as chemical validity and structural similarity. Here we introduce Minimum Risk Training (MRT) to OCSR and propose COMO (Closed-loop Optical Molecule recOgnition), a closed-loop framework that mitigates exposure bias by directly optimizing over molecule-level, non-differentiable objectives, by iteratively sampling and evaluating the model's own predictions. Experiments on ten benchmarks including synthetic and real-world chemical diagrams from patent and scientific literature demonstrate that COMO substantially outperforms existing rule-based and learning-based methods with less training data. Ablation studies further show that MRT is architecture-agnostic, demonstrating its potential for broad application to end-to-end OCSR systems.

Abstract

Optical Character Recognition

Explore similar work

COMO: Closed-Loop Optical Molecule Recognition with Minimum Risk Training

MolRecBench-Wild: A Real-World Benchmark for Optical Chemical Structure Recognition

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