Boosting Document Parsing Efficiency and Performance with Coarse-to-Fine Visual Processing

Cheng Cui; Ting Sun; Suyin Liang; Tingquan Gao; Zelun Zhang; Jiaxuan Liu; Xueqing Wang; Changda Zhou

cs.CVMar 25, 2026

Boosting Document Parsing Efficiency and Performance with Coarse-to-Fine Visual Processing

Authors: Cheng Cui, Ting Sun, Suyin Liang, Tingquan Gao, Zelun Zhang, Jiaxuan Liu, Xueqing Wang, Changda Zhou,

Organizations: PaddlePaddle Team, Baidu Inc. · Xi’an Jiaotong University

Abstract

Document parsing is a fine-grained task where image resolution significantly impacts performance. While advanced research leveraging vision-language models benefits from high-resolution input to boost model performance, this often leads to a quadratic increase in the number of vision tokens and significantly raises computational costs. We attribute this inefficiency to substantial visual regions redundancy in document images, like background. To tackle this, we propose PaddleOCR-VL, a novel coarse-to-fine architecture that focuses on semantically relevant regions while suppressing redundant ones, thereby improving both efficiency and performance. Specifically, we introduce a lightweight Valid Region Focus Module (VRFM) which leverages localization and contextual relationship prediction capabilities to identify valid vision tokens. Subsequently, we design and train a compact yet powerful 0.9B vision-language model (PaddleOCR-VL-0.9B) to perform detailed recognition, guided by VRFM outputs to avoid direct processing of the entire large image. Extensive experiments demonstrate that PaddleOCR-VL achieves state-of-the-art performance in both page-level parsing and element-level recognition. It significantly outperforms existing solutions, exhibits strong competitiveness against top-tier VLMs, and delivers fast inference while utilizing substantially fewer vision tokens and parameters, highlighting the effectiveness of targeted coarse-to-fine parsing for accurate and efficient document understanding. The source code and models are publicly available at https://github.com/PaddlePaddle/PaddleOCR.

Figures & tables

Figure 1: PaddleOCR-VL achieves state-of-the-art performance with the fewest vision tokens and parameters on OmniDocBench v1.5.

Figure 2: Architectural comparison of End-to-end VLM and our method. Among various types of document images, the valid area accounts for less than 50% on average. Therefore, our proposed PaddleOCR-VL can achieve improvements in visual information processing by discarding redundant regions.

Figure 3: The overview of our proposed PaddleOCR-VL, which consists of two components. The first component is VRFM, which accurately extracts valid regions containing layout elements from unstructured document images and predicts the reading order. Subsequently, each valid region is fed into our designed PaddleOCR-VL-0.9B for precise recognition of individual elements. The final results are then reconstructed into a structured document according to the reading order.

Figure 4: Architecture of the Valid Region Focus Module (VRFM).

Stages	Stage 1	Stage 2
Training Samples	29M	2.7M
Max Resolution	1280 $\times$ 28 $\times$ 28	2048 $\times$ 28 $\times$ 28
Sequence length	16384	16384
Trainable components	All	All
Batch sizes	128	128
Data Augmentation	Yes	Yes

Table 1: Training settings of PaddleOCR-VL-0.9B.

Model Type	Methods	Parameters	Vision Tokens	Overall $\uparrow$	Text Edit $\downarrow$	Formula CDM $\uparrow$	Table TEDS $\uparrow$	Table TEDS-S $\uparrow$	Reading Order Edit $\downarrow$
Pipeline Tools	Marker-1.8.2 [ 44 ]	-	-	71.30	0.206	76.66	57.88	71.17	0.250
	Mineru2-pipeline [ 40 ]	-	-	75.51	0.209	76.55	70.90	79.11	0.225
	PP-StructureV3 [ 9 ]	-	-	86.73	0.073	85.79	81.68	89.48	0.073
General VLMs	GPT-4o [ 1 ]	-	-	75.02	0.217	79.70	67.07	76.09	0.148
	InternVL3-76B [ 63 ]	76B	7277	80.33	0.131	83.42	70.64	77.74	0.113
	InternVL3.5-241B [ 53 ]	241B	7277	82.67	0.142	87.23	75.00	81.28	0.125

Table 2: Comprehensive evaluation of document parsing on OmniDocBench v1.5. Results are reported by OmniDocBench [ 41 ] unless Ours. In PaddleOCR-VL, S, M, and L refer to the same PaddleOCR-VL weights using different vsion tokens.

Methods	Overall TEDS $\uparrow$	Structural TEDS $\uparrow$	Overall Edit Dist $\downarrow$
MinerU2-VLM [ 40 ]	0.9002	0.9369	0.0734
dots.ocr [ 46 ]	0.8194	0.8442	0.1508
MinerU2.5 [ 39 ]	0.9005	0.9539	0.0693
PaddleOCR-VL-L	0.9046	0.9420	0.0681

Table 3: Comparison of OmniDocBench-Table-block Performance

Methods	Edit Distance $\downarrow$
Methods	PPT	Academic Literature	Book	Colorful Textbook	Exam Paper	Magazine	Newspaper	Note	Research Report
Qwen2.5-VL-72B [ 3 ]	0.054	0.023	0.061	0.084	0.195	0.032	0.056	0.118	0.040
MonkeyOCR-pro-3B [ 23 ]	0.058	0.021	0.064	0.096	0.116	0.023	0.058	0.124	0.052
MinerU2.5 [ 39 ]	0.195	0.089	0.111	0.234	0.194	0.147	0.056	0.142	0.094
Dolphin [ 13 ]	0.237	0.095	0.135	0.347	0.248	0.233	0.121	0.309	0.213
PaddleOCR-VL-L	0.049	0.021	0.047	0.082	0.115	0.020	0.035	0.077	0.031

Table 4: Overall Comparison of OmniDocBench-OCR-block Performance.

Methods	Overall CDM $\uparrow$	EN CDM $\uparrow$	ZH CDM $\uparrow$
dots.ocr [ 46 ]	0.4641	0.4868	0.4414
MinerU2-VLM [ 40 ]	0.8286	0.9616	0.6956
MonkeyOCR-pro-1.2B [ 23 ]	0.8531	0.9642	0.7419
MonkeyOCR-3B [ 23 ]	0.8621	0.9718	0.7524
Qwen2.5-VL-72B [ 3 ]	0.8747	0.9574	0.7920
MinerU2.5 [ 39 ]	0.9187	0.9751	0.8623

Table 5: Comparison of OmniDocBench v1.5 Formula-block Performance. Due to dots.ocr [ 46 ] easily recognizing cropped formulas as images, the score is relatively low.

Methods	RMS-F1 $\uparrow$
Methods	Overall	EN	ZH
TinyChart [ 59 ]	0.2159	0.4726	0.0876
GOT [ 54 ]	0.3160	0.1100	0.4190
OneChart [ 7 ]	0.3716	0.1384	0.4882
Qwen2.5-VL-72B [ 3 ]	0.7300	0.6972	0.7464
PP-StructureV3 [ 9 ]	0.8060	0.7963	0.8109

Table 6: Comparison of In-house-Chart Performance

Methods	Total Time (s) $\downarrow$	Pages/s $\uparrow$	Tokens/s $\uparrow$	Avg. VRAM Usage (GB) $\downarrow$
MonkeyOCR-pro-1.2B [ 23 ]	1456.4	0.6730	1120.3	75.5
dots.ocr [ 46 ]	2784.6	0.3522	532.9	78.5
MinerU2.5 [ 39 ]	927.3	1.0574	1647.9	41.9
PaddleOCR-VL-L	605.2	1.6192	2470.7	42.1

Table 7: End-to-End Inference Performance Comparison. All models were tested using the vLLM 0.10.2 backend.

Figure S1 : The construction method and characteristics of the text training data for PaddleOCR-VL-0.9B.

Figure S2 : The construction method and characteristics of the table training data for PaddleOCR-VL-0.9B.

Figure S3 : The construction method and characteristics of the formula training data for PaddleOCR-VL-0.9B.

Figure S4 : The construction method and characteristics of the chart training data for PaddleOCR-VL-0.9B.

Language Category	Specific Languages
Chinese	Chinese
English	English
Korean	Korean
Japanese	Japanese
Thai	Thai
Greek	Greek

Table S1 : Supported Languages

Method	Visual Tokens	Acc.(%)
Qwen3-VL-8B [ 2 ]	5818	89.60
DocKylin [ 58 ] + Qwen3-VL-8B [ 2 ]	4316	87.75
PP-DocLayout [ 47 ] + Qwen3-VL-8B	3876	89.30
PP-DocLayout + PaddleOCR-VL-0.9B	2347	89.96
VRFM + Qwen3-VL-8B	4262	90.37
VRFM + PaddleOCR-VL-0.9B	2561	92.62

Table S2: Comparison of VRFM with representative baselines in terms of token efficiency and accuracy.

Language	Qwen2.5-VL-72B	Dolphin	MinerU2.5	Ours
Arabic	0.405	0.682	0.978	0.122
Korean	0.056	0.699	0.917	0.052
Tamil	0.389	0.912	0.957	0.043
Greek	0.165	0.691	0.661	0.135
Thai	0.194	0.709	0.880	0.081
Telugu	0.758	0.832	0.937	0.011

Table S3: Multilingual OCR performance comparison (Edit Distance

\downarrow

Hardware	Backend	Total Time (s) $\downarrow$	Pages/s $\uparrow$	Tokens/s $\uparrow$	Avg. VRAM Usage (GB) $\downarrow$
A100	vLLM	800.9	1.2241	1881.2	43.7
A100	SGLang	917.6	1.0684	1641.5	49.8
A10	vLLM	1238.0	0.7921	1217.2	14.1
A10	SGLang	1429.9	0.6858	1055.8	20.0
RTX 3060	vLLM	2749.1	0.3568	548.2	11.9
RTX 3060	SGLang	2792.4	0.3513	540.8	11.8

Table S4 : End-to-End Inference Performance

Figure S5 : The Layout and Markdown Output for Book, Textbook and Academic Paper.

Figure S6 : The Layout and Markdown Output for Research Report(with chart recognition enabled), Financial Report, Slides and Exam Paper.

Figure S7 : The Layout and Markdown Output for Notes, Vertical Book and Ancient Book.

Figure S8 : The Layout and Markdown Output for Certificate, Newspaper and Magazine.

Figure S9 : The Layout Detection results for various types of documents.

Figure S10 : The Layout Detection results for various types of documents.

Figure S11 : The Layout Detection results for various types of documents.

Figure S12 : The Reading Order results for various types of documents.

Figure S13 : The Reading Order results for various types of documents.

Figure S14 : The markdown output for French and Hindi documents.

Figure S15 : The markdown output for Croatian and Spanish documents.

Figure S16 : The markdown output for English and Arabic documents.

Figure S17 : The markdown output for German and Chinese documents.

Figure S18 : The markdown output for Russian and Japanese documents.

Figure S19 : The markdown output for Thai and Korean documents.

Figure S20 : The markdown output for Mixed Printed Handwritten Text and Handwritten Formula documents.

Figure S21 : The markdown output for Handwriting Chinese and Handwriting English documents.

Figure S22 : The markdown output for various types of vertical documents.

Figure S23 : The markdown output for various types of Tables.

Figure S24 : The markdown output for various types of Tables.

Figure S25 : The markdown output for various types of Formulas.

Figure S26 : The markdown output for various types of Formulas.

Figure S27 : The markdown output for various types of Charts.

Figure S28 : The markdown output for various types of Charts.

Figure S29 : The markdown output for various types of Charts.

Figure S30 : Compare with others in Layout Detection.

Figure S31 : Compare with others in Layout Detection.

Figure S32 : Compare with others in Multilingual Text Recognition.

Figure S33 : Compare with others in Multilingual Text Recognition.

Figure S34 : Compare with others in Multilingual Text Recognition.

Figure S35 : Compare with others in Handwriting Text Recognition.

Figure S36 : Compare with others in Handwriting Text Recognition.

Figure S37 : Compare with others in Vertical Text Recognition.

Figure S38 : Compare with others in Table Recognition.

Figure S39 : Compare with others in Table Recognition.

Figure S40 : Compare with others in Formula Recognition.

Figure S41 : Compare with others in Chart Recognition.

Figure S42 : Compare with others in Chart Recognition.

Document Parsing

Document parsing converts PDFs and page images into structured, machine-readable text, tables, formulas, layouts, and reading order.

See all 63 papers

PaddleOCR-VL-1.6: Expanding the Frontier of Document Parsing with Under-Optimized Region Refinement and Progressive Post-Training
Jun 2, 2026Zelun Zhang, Hongen Liu, Suyin Liang +12Data Selection Training Data Selection
PaddlePaddle Team, Baidu Inc.
Efficient Document Parsing via Parallel Token Prediction
Mar 16, 2026Lei Li, Ze Zhao, Meng Li +6Efficient VLM Inference Document Parsing
Platform and Content Group, Tencent · Renmin University of China
HSD: Training-Free Acceleration for Document Parsing Vision-Language Models with Hierarchical Speculative Decoding
Feb 13, 2026Wenhui Liao, Hongliang Li, Pengyu Xie +15Efficient VLM Inference Speculative Decoding
South China University of Technology, Guangzhou, China · Shanghai Artificial Intelligence Laboratory, Shanghai, China · Nanjing University, Nanjing, China +1

Jun 2, 2026J/K move · Enter open · S save

PaddleOCR-VL-1.6: Expanding the Frontier of Document Parsing with Under-Optimized Region Refinement and Progressive Post-Training

Zelun Zhang, Hongen Liu, Suyin Liang, Yubo Zhang+11

PaddlePaddle Team, Baidu Inc.

Data Selection Training Data Selection Document Parsing Document Image Understanding

Open paper AskPDF arXiv

We introduce PaddleOCR-VL-1.6, an upgraded compact document parsing model built upon PaddleOCR-VL-1.5. Although PaddleOCR-VL-1.5 establishes a strong 0.9B baseline, its remaining errors concentrate in under-optimized regions where model behavior is unstable, data coverage is sparse, or supervision is unreliable. Rather than expanding the training corpus indiscriminately, PaddleOCR-VL-1.6 introduces a region-aware data optimization framework that identifies weak regions from the previous model, applies targeted enhancement to these regions, and improves the reliability of supervision signals. It further adopts a progressive post-training recipe based on curated data selection and reinforcement learning, pushing model performance to a higher level through staged optimization. PaddleOCR-VL-1.6 achieves a new state-of-the-art score of 96.33% on OmniDocBench v1.6, demonstrates strong competitiveness against top-tier VLMs, and provides a practical post-training recipe for the PaddleOCR-VL series.

Abstract

Document Parsing

Explore similar work

PaddleOCR-VL-1.6: Expanding the Frontier of Document Parsing with Under-Optimized Region Refinement and Progressive Post-Training

Efficient Document Parsing via Parallel Token Prediction

HSD: Training-Free Acceleration for Document Parsing Vision-Language Models with Hierarchical Speculative Decoding