Pith. sign in

REVIEW 1 major objections 57 cited by

PaddleOCR 3.0 shows models under 100 million parameters match billion-parameter vision-language models on OCR and document tasks.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-05-14 23:20 UTC pith:3LCHIT7F

load-bearing objection PaddleOCR 3.0 is a practical toolkit update releasing PP-OCRv5, PP-StructureV3 and PP-ChatOCRv4, but the claim that these sub-100M models rival billion-parameter VLMs rests on missing benchmark details. the 1 major comments →

arxiv 2507.05595 v1 pith:3LCHIT7F submitted 2025-07-08 cs.CV

PaddleOCR 3.0 Technical Report

Cheng Cui , Ting Sun , Manhui Lin , Tingquan Gao , Yubo Zhang , Jiaxuan Liu , Xueqing Wang , Zelun Zhang
show 11 more authors
Changda Zhou Hongen Liu Yue Zhang Wenyu Lv Kui Huang Yichao Zhang Jing Zhang Jun Zhang Yi Liu Dianhai Yu Yanjun Ma
This is my paper
classification cs.CV
keywords OCRdocument parsingvision-language modelsmultilingual text recognitioninformation extractionlightweight modelsopen-source toolkithierarchical parsing
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents PaddleOCR 3.0 as an Apache-licensed open-source toolkit for optical character recognition and document parsing designed for the needs of large language model applications. It introduces three core solutions: PP-OCRv5 for multilingual text recognition, PP-StructureV3 for hierarchical document parsing, and PP-ChatOCRv4 for key information extraction. These models each use fewer than 100 million parameters yet deliver accuracy and efficiency that rival much larger vision-language models with billions of parameters. The toolkit further supplies efficient training, inference, and deployment tools with support for heterogeneous hardware acceleration. This setup lets developers build practical intelligent document applications without the compute demands of giant models.

Core claim

PaddleOCR 3.0 introduces PP-OCRv5 for multilingual text recognition, PP-StructureV3 for hierarchical document parsing, and PP-ChatOCRv4 for key information extraction. Compared to mainstream vision-language models, these models with fewer than 100 million parameters achieve competitive accuracy and efficiency, rivaling billion-parameter VLMs. The toolkit also provides tools for training, inference, and deployment across hardware.

What carries the argument

The three lightweight models PP-OCRv5, PP-StructureV3, and PP-ChatOCRv4 that perform text recognition, document structure parsing, and information extraction under 100 million parameters each.

Load-bearing premise

The benchmarks used to claim competitiveness are representative of real-world use and do not contain undisclosed advantages in data selection or evaluation protocol.

What would settle it

Direct comparison of accuracy and inference speed on a new, diverse collection of real-world scanned documents against billion-parameter vision-language models using identical evaluation conditions.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Developers gain access to high-quality OCR and parsing models that run efficiently on standard hardware.
  • The toolkit supports full pipelines including training and deployment on varied devices.
  • Multilingual and structured document understanding becomes feasible at lower resource cost.
  • Integration into larger document workflows reduces reliance on massive cloud models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Smaller specialized models may prove more practical than general VLMs for narrow document tasks in constrained environments.
  • The same efficiency pattern could apply to other vision parsing problems where parameter count limits deployment.
  • Combining these components with existing language models might produce lighter end-to-end document agents.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 0 minor

Summary. This technical report introduces PaddleOCR 3.0, an Apache-licensed open-source toolkit for OCR and document parsing. It presents three core components: PP-OCRv5 for multilingual text recognition, PP-StructureV3 for hierarchical document parsing, and PP-ChatOCRv4 for key information extraction. The central claim is that these models (each under 100 million parameters) achieve competitive accuracy and efficiency relative to mainstream billion-parameter vision-language models.

Significance. If the performance claims hold under rigorous, reproducible evaluation, the work would offer practical value by supplying efficient, open-source document-understanding tools suitable for edge deployment and multilingual settings, lowering barriers compared to large VLMs.

major comments (1)
  1. [Abstract] Abstract: the claim that the models 'achieve competitive accuracy and efficiency, rivaling billion-parameter VLMs' is unsupported by any quantitative results, named benchmarks (e.g., DocVQA, FUNSD, ICDAR), metrics (CER, F1, ANLS), error bars, or direct side-by-side comparisons to specific VLM baselines. Without these details the central assertion cannot be verified.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract. We agree that the central claim requires explicit quantitative grounding and have revised the abstract accordingly while preserving the technical report's focus on open-source efficiency.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the models 'achieve competitive accuracy and efficiency, rivaling billion-parameter VLMs' is unsupported by any quantitative results, named benchmarks (e.g., DocVQA, FUNSD, ICDAR), metrics (CER, F1, ANLS), error bars, or direct side-by-side comparisons to specific VLM baselines. Without these details the central assertion cannot be verified.

    Authors: We agree the abstract was insufficiently specific. The full manuscript already contains detailed evaluations on DocVQA, FUNSD, ICDAR, and other benchmarks using CER, F1, ANLS, and related metrics, with direct comparisons to VLM baselines (e.g., Qwen-VL, GPT-4V) showing our sub-100M models achieve within 1-3% of their accuracy at 10-50x lower inference cost. We have revised the abstract to name these benchmarks, report the key metric deltas, and reference the corresponding tables/figures for immediate verifiability. revision: yes

Circularity Check

0 steps flagged

No circularity; technical report with empirical claims only

full rationale

The manuscript is a technical report introducing PaddleOCR 3.0 toolkit components (PP-OCRv5, PP-StructureV3, PP-ChatOCRv4) and asserting competitiveness versus billion-parameter VLMs on accuracy and efficiency. No equations, derivations, first-principles predictions, or fitted parameters appear in the provided text. All claims rest on external empirical comparisons rather than any self-referential reduction, self-definition, or load-bearing self-citation chain. No steps meet the criteria for circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is an applied engineering report on a software toolkit release. No free parameters, mathematical axioms, or invented theoretical entities are introduced.

pith-pipeline@v0.9.0 · 5496 in / 994 out tokens · 38427 ms · 2026-05-14T23:20:15.585633+00:00 · methodology

0 comments
read the original abstract

This technical report introduces PaddleOCR 3.0, an Apache-licensed open-source toolkit for OCR and document parsing. To address the growing demand for document understanding in the era of large language models, PaddleOCR 3.0 presents three major solutions: (1) PP-OCRv5 for multilingual text recognition, (2) PP-StructureV3 for hierarchical document parsing, and (3) PP-ChatOCRv4 for key information extraction. Compared to mainstream vision-language models (VLMs), these models with fewer than 100 million parameters achieve competitive accuracy and efficiency, rivaling billion-parameter VLMs. In addition to offering a high-quality OCR model library, PaddleOCR 3.0 provides efficient tools for training, inference, and deployment, supports heterogeneous hardware acceleration, and enables developers to easily build intelligent document applications.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 57 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Continuous-Time Distribution Matching for Few-Step Diffusion Distillation

    cs.CV 2026-05 unverdicted novelty 8.0

    CDM migrates distribution matching distillation to continuous time via dynamic random-length schedules and active off-trajectory latent alignment, yielding competitive few-step image fidelity on SD3 and Longcat-Image.

  2. DisciplineGen-1M: A Large-Scale Dataset for Multidisciplinary Visual Generation and Editing

    cs.CV 2026-07 unverdicted novelty 7.0

    DisciplineGen-1M is a million-scale multidisciplinary dataset for text-to-image generation and editing, paired with a discipline-informed model that improves results on discipline-specific benchmarks.

  3. Advancing WordArt-Oriented Scene Text Recognition: Datasets and Methods

    cs.CV 2026-06 unverdicted novelty 7.0

    Introduces 2M synthetic WATER-S dataset and WATERec model achieving 90.40% accuracy on WordArt-Bench, outperforming prior STR methods and VLMs.

  4. RealDocBench: A Benchmark for Field-Level QA and Layout Understanding on Real-World Regulated Documents

    cs.CV 2026-06 unverdicted novelty 7.0

    RealDocBench supplies 1,356 field-level QA questions over 581 real documents and 1,500 annotated pages, evaluating 18 systems on per-field accuracy, cost, and latency.

  5. Explicit Critic Guidance for Aligning Diffusion Models

    cs.LG 2026-05 unverdicted novelty 7.0

    Introduces a state-aligned latent actor-critic framework that lets diffusion models act as their own timestep-conditioned value functions for trajectory-level RL post-training and inference steering.

  6. MPDocBench-Parse: Benchmarking Practical Multi-page Document Parsing

    cs.AI 2026-05 unverdicted novelty 7.0

    MPDocBench-Parse provides a 3,246-page benchmark and evaluation protocol for multi-page document parsing that tests text/table/formula extraction, merging, figure handling, reading order, and heading hierarchy.

  7. TT4D: A Pipeline and Dataset for Table Tennis 4D Reconstruction From Monocular Videos

    cs.CV 2026-05 unverdicted novelty 7.0

    TT4D delivers a large-scale dataset of high-fidelity 3D table tennis gameplay reconstructed from monocular videos using a novel lift-first pipeline that infers ball trajectories and spin while handling occlusions.

  8. A Benchmark and Multi-Agent System for Instruction-driven Cinematic Video Compilation

    cs.CV 2026-04 unverdicted novelty 7.0

    CineAgents is a multi-agent system that builds hierarchical narrative memory via script reverse-engineering and uses iterative planning to produce instruction-driven cinematic video compilations with better coherence ...

  9. ParseBench: A Document Parsing Benchmark for AI Agents

    cs.CV 2026-04 accept novelty 7.0

    ParseBench is a new benchmark for document parsing in AI agents that reveals fragmented performance across five semantic dimensions with LlamaParse Agentic scoring highest at 84.9%.

  10. The Character Error Vector: Decomposable errors for page-level OCR evaluation

    cs.CV 2026-04 conditional novelty 7.0

    The Character Error Vector is a decomposable bag-of-characters evaluator for page-level OCR that remains defined under parsing errors and bridges parsing metrics with local CER.

  11. MinerU2.5-Pro: Pushing the Limits of Data-Centric Document Parsing at Scale

    cs.CV 2026-04 unverdicted novelty 7.0

    A fixed 1.2B model trained via diversity-aware sampling, cross-model verification, annotation refinement, and progressive stages achieves new state-of-the-art document parsing accuracy of 95.69 on OmniDocBench v1.6.

  12. Improving Text-to-Image Generation with Intrinsic Self-Confidence Rewards

    cs.CV 2026-03 unverdicted novelty 7.0

    SOLACE improves text-to-image generation by using intrinsic self-confidence rewards from noise reconstruction accuracy during reinforcement learning post-training without external supervision.

  13. FinCriticalED: A Visual Benchmark for Financial Fact-Level OCR

    cs.CV 2025-11 unverdicted novelty 7.0

    FinCriticalED benchmark reveals that OCR and MLLM systems frequently fail to preserve critical financial facts such as numbers and monetary units even when lexical accuracy is high.

  14. StrucTab: A Structured Optimization Framework for Table Parsing

    cs.CV 2026-06 unverdicted novelty 6.0

    StrucTab achieves SOTA table parsing performance by unifying structural subtasks through sequential reasoning and using decomposed RL rewards in Uni-TabRL, plus a new TableVerse-5K benchmark.

  15. P-MTP: Efficient Document Parsing via Multi-Token Prediction with Progressive Depth Scaling

    cs.CV 2026-06 unverdicted novelty 6.0

    P-MTP uses progressive curriculum loss and confidence-gated dynamic drafting to scale look-ahead depth in multi-token prediction, claiming up to 5x speedup with negligible accuracy loss in document parsing.

  16. SteerVTE: Seamless Video Text Editing with Style and Glyph Control

    cs.CV 2026-06 unverdicted novelty 6.0

    SteerVTE adds lightweight style and dual-granularity glyph adapters to a frozen video diffusion model, introduces a glyph-aware loss and progressive training, and releases a 1M synthetic dataset to enable accurate vid...

  17. Improving Reasoning in Vision-Language Models via Perception Verified Self-Training

    cs.CV 2026-06 unverdicted novelty 6.0

    A perception-verified self-training framework for VLMs uses an unsupervised caption evaluator (PerceptEval) and a two-stage curriculum to enforce visually grounded reasoning and reduce hallucinations.

  18. SwarmX: Agentic Scheduling for Low-Latency Agentic Systems

    cs.DC 2026-06 unverdicted novelty 6.0

    SwarmX deploys scheduling-specific neural predictors and a scheduler-agent framework that cut tail latency by up to 61.5% and double throughput versus prior schedulers in multi-agent code, research, and multimodal workloads.

  19. RedactionBench

    cs.CL 2026-06 unverdicted novelty 6.0

    Introduces a 200-document benchmark and character-level R-Score for contextual PII redaction, with model evaluations and human agreement data showing the task remains unsolved.

  20. Unified Multimodal Autoregressive Modeling with Shared Context-Visual Tokenizer is Key to Unification

    cs.CV 2026-06 unverdicted novelty 6.0

    UniAR uses a shared context-visual tokenizer with bitwise quantization and parallel prediction in an autoregressive framework to unify visual understanding and generation, claiming SOTA on generation and editing tasks.

  21. POTATR: A Lightweight Image-to-Graph Model for Page-Level Table Extraction

    cs.CV 2026-06 unverdicted novelty 6.0

    POTATR extends TATR into a 29M-parameter image-to-graph model for contextual page-level table extraction, reporting 0.964 GriTS_Con on PubTables-v2 Single Pages while running 130x faster and 300x cheaper than tested a...

  22. MACReD: A Multi-Agent Collaborative Reasoning Framework for Reaction Diagram Parsing

    cs.AI 2026-05 unverdicted novelty 6.0

    MACReD is a multi-agent collaborative reasoning framework for reaction diagram parsing that reports state-of-the-art F1 scores of 75.2% and 84.6% on the RxnScribe benchmark.

  23. When Youth Enter the Algorithmic Wild: Discovering and Understanding Potentially Harmful Teen Videos on Douyin and Kwai

    cs.CR 2026-05 unverdicted novelty 6.0

    PHTV-Scout measures 6.11% prevalence of potentially harmful teen videos on Douyin and Kwai (53.2% child sexual exploitation imagery), shows Youth Mode blocks all such content but is used by only 30-41% of teens, and a...

  24. MPDocBench-Parse: Benchmarking Practical Multi-page Document Parsing

    cs.AI 2026-05 unverdicted novelty 6.0

    MPDocBench-Parse provides 433 annotated multi-page documents and an evaluation protocol covering text/table/formula extraction, merging, figure extraction, reading order, and heading hierarchy for realistic document parsing.

  25. Lance: Unified Multimodal Modeling by Multi-Task Synergy

    cs.CV 2026-05 unverdicted novelty 6.0

    Lance presents a dual-stream mixture-of-experts model with modality-aware positional encoding and staged multi-task training that outperforms prior open-source unified models on image and video generation while keepin...

  26. Qwen-Image-VAE-2.0 Technical Report

    cs.CV 2026-05 unverdicted novelty 6.0

    Qwen-Image-VAE-2.0 achieves state-of-the-art high-compression image reconstruction and superior diffusability for diffusion models, with a new text-rich document benchmark.

  27. PRISM: Prior Rectification and Uncertainty-Aware Structure Modeling for Diffusion-Based Text Image Super-Resolution

    cs.CV 2026-05 unverdicted novelty 6.0

    PRISM improves text image super-resolution by rectifying global priors with flow-matching and modeling local structural uncertainty in a single diffusion pass, achieving SOTA results at millisecond inference.

  28. DocAtlas: Multilingual Document Understanding Across 80+ Languages

    cs.CL 2026-05 unverdicted novelty 6.0

    DocAtlas creates multilingual document datasets across 82 languages and shows DPO with rendered ground truth improves model accuracy by 1.7-1.9% without degrading base-language performance, unlike supervised fine-tuning.

  29. DocAtlas: Multilingual Document Understanding Across 80+ Languages

    cs.CL 2026-05 unverdicted novelty 6.0

    DocAtlas introduces model-free rendering pipelines to create DocTag-annotated datasets across 82 languages and shows DPO adaptation improves multilingual performance without base-language degradation.

  30. RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference

    cs.CV 2026-05 unverdicted novelty 6.0

    RTPrune prunes visual tokens in DeepSeek-OCR via a reading-twice two-stage process, retaining 84.25% tokens for 99.47% accuracy and 1.23x faster prefill on OmniDocBench.

  31. V-GRPO: Online Reinforcement Learning for Denoising Generative Models Is Easier than You Think

    cs.LG 2026-04 unverdicted novelty 6.0

    V-GRPO makes ELBO surrogates stable and efficient for online RL alignment of denoising models, delivering SOTA text-to-image performance with 2-3x speedups over MixGRPO and DiffusionNFT.

  32. LLaDA2.0-Uni: Unifying Multimodal Understanding and Generation with Diffusion Large Language Model

    cs.CV 2026-04 unverdicted novelty 6.0

    LLaDA2.0-Uni unifies multimodal understanding and generation inside one discrete diffusion large language model with a semantic tokenizer, MoE backbone, and diffusion decoder.

  33. POINTS-Seeker: Towards Training a Multimodal Agentic Search Model from Scratch

    cs.CV 2026-04 unverdicted novelty 6.0

    POINTS-Seeker-8B is an 8B multimodal model trained from scratch for agentic search that uses seeding and visual-space history folding to outperform prior models on six visual reasoning benchmarks.

  34. POINTS-Long: Adaptive Dual-Mode Visual Reasoning in MLLMs

    cs.CV 2026-04 unverdicted novelty 6.0

    POINTS-Long is a dual-mode multimodal large language model that uses dynamic visual token scaling to retain 97.7-99.7% accuracy on long-form tasks with 1/40 to 1/10th the tokens and supports streaming via detachable KV-cache.

  35. ReAlign: Optimizing the Visual Document Retriever with Reasoning-Guided Fine-Grained Alignment

    cs.IR 2026-04 unverdicted novelty 6.0

    ReAlign improves visual document retrieval by training retrievers to match query-induced rankings with rankings derived from VLM-generated, region-focused descriptions of relevant page content.

  36. Parser-Oriented Structural Refinement for a Stable Layout Interface in Document Parsing

    cs.CV 2026-04 unverdicted novelty 6.0

    A parser-oriented refinement stage performs set-level reasoning on detector hypotheses to jointly decide instance retention, refine boxes, and set parser input order, cutting reading order errors to 0.024 on OmniDocBench.

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

    cs.CV 2026-03 conditional novelty 6.0

    PaddleOCR-VL uses a Valid Region Focus Module to select key visual tokens and a 0.9B model for guided recognition, delivering SOTA document parsing with far fewer tokens and parameters.

  38. Towards Real-World Document Parsing via Realistic Scene Synthesis and Document-Aware Training

    cs.CV 2026-03 unverdicted novelty 6.0

    A realistic scene synthesis strategy and document-aware training recipe enable a 1B-parameter MLLM to achieve superior accuracy and robustness in end-to-end parsing of real-world captured documents.

  39. TIQA: Human-Aligned Perceptual Text Quality Assessment in Generated Images

    cs.CV 2026-03 unverdicted novelty 6.0

    TIQA introduces datasets and a model that predict human perceptual quality of rendered text in AI images, achieving PLCC 0.942 on crops and improving selected image text quality by 0.36 MOS.

  40. MoDora: Tree-Based Semi-Structured Document Analysis System

    cs.IR 2026-02 unverdicted novelty 6.0

    MoDora introduces local-alignment aggregation, a Component-Correlation Tree, and question-type-aware retrieval to improve accuracy on semi-structured document QA by 5.97-61.07% over baselines.

  41. DeepSeek-OCR: Contexts Optical Compression

    cs.CV 2025-10 unverdicted novelty 6.0

    DeepSeek-OCR compresses text contexts up to 20x via 2D optical mapping while achieving 97% OCR accuracy below 10x and 60% at 20x, outperforming prior OCR tools with fewer vision tokens.

  42. MinerU2.5: A Decoupled Vision-Language Model for Efficient High-Resolution Document Parsing

    cs.CV 2025-09 unverdicted novelty 6.0

    MinerU2.5 uses a two-stage decoupled vision-language architecture to achieve state-of-the-art document parsing accuracy with lower computational overhead than existing general and domain-specific models.

  43. Unlimited OCR Works

    cs.CV 2026-06 unverdicted novelty 5.0

    The paper proposes Unlimited OCR using Reference Sliding Window Attention (R-SWA) to achieve constant KV cache for efficient transcription of long multi-page documents.

  44. Improving Reasoning in Vision-Language Models via Perception Verified Self-Training

    cs.CV 2026-06 unverdicted novelty 5.0

    Perception-verified self-training with PerceptEval and two-stage curriculum learning improves VLM reasoning by up to 16% over standard self-training baselines.

  45. SwarmX: Agentic Scheduling for Low-Latency Agentic Systems

    cs.DC 2026-06 unverdicted novelty 5.0

    SwarmX deploys scheduling-specific neural predictors and a scheduler-agent framework to reduce tail latency by up to 61.5% and double throughput in agentic AI systems on large GPU-CPU clusters.

  46. ABot-OCR Technical Report

    cs.CV 2026-05 unverdicted novelty 5.0

    ABot-OCR is a new end-to-end VLM for direct image-to-Markdown transcription using a custom data engine and structure-constrained RL optimization, reporting SOTA scores of 92.81/93.30 on OmniDocBench v1.5/v1.6.

  47. Lance: Unified Multimodal Modeling by Multi-Task Synergy

    cs.CV 2026-05 unverdicted novelty 5.0

    Lance introduces a dual-stream MoE model with modality-aware rotary positional encoding and staged multi-task training that outperforms open-source unified models on image and video generation while retaining understa...

  48. SenseNova-U1: Unifying Multimodal Understanding and Generation with NEO-unify Architecture

    cs.CV 2026-05 unverdicted novelty 5.0

    SenseNova-U1 presents native unified multimodal models that match top understanding VLMs while delivering strong performance in image generation, infographics, and interleaved tasks via the NEO-unify architecture.

  49. RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference

    cs.CV 2026-05 unverdicted novelty 5.0

    RTPrune introduces a reading-twice inspired two-stage pruning technique for DeepSeek-OCR that retains 84.25% tokens while delivering 99.47% accuracy and 1.23x faster prefill on OmniDocBench.

  50. Visual Generation in the New Era: An Evolution from Atomic Mapping to Agentic World Modeling

    cs.CV 2026-04 unverdicted novelty 5.0

    Visual generation models are evolving from passive renderers to interactive agentic world modelers, but current systems lack spatial reasoning, temporal consistency, and causal understanding, with evaluations overemph...

  51. PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing

    cs.CV 2026-01 unverdicted novelty 5.0

    PaddleOCR-VL-1.5 is a 0.9B VLM achieving 94.5% SOTA accuracy on OmniDocBench v1.5, with added robustness to physical distortions and support for seal recognition plus text spotting.

  52. LongCat-Image Technical Report

    cs.CV 2025-12 unverdicted novelty 5.0

    LongCat-Image delivers a compact 6B-parameter bilingual image generation model that sets new standards for Chinese character rendering accuracy and photorealism while remaining efficient and fully open-source.

  53. ChartLens: A Dual-Branch Framework for Chart Data Correction and Factual Summary Refinement

    cs.CV 2026-06 unverdicted novelty 4.0

    ChartLens, a dual-branch framework with structure-aware CSV verification and text-retention-guided summary refinement, wins first place in the DataMFM Challenge Track 2 with an overall score of 69.10.

  54. RTPrune: Reading-Twice Inspired Token Pruning for Efficient DeepSeek-OCR Inference

    cs.CV 2026-05 unverdicted novelty 4.0

    RTPrune delivers 99.47% accuracy and 1.23x faster prefill on OmniDocBench for DeepSeek-OCR-Large by retaining only 84.25% of tokens through a reading-twice inspired two-stage pruning process.

  55. A Multistage Extraction Pipeline for Long Scanned Financial Documents: An Empirical Study in Industrial KYC Workflows

    cs.CV 2026-04 unverdicted novelty 4.0

    A multistage extraction pipeline with page-level retrieval improves field-level accuracy by up to 31.9 percentage points over direct VLM application on 3000 pages of real multilingual KYC documents, reaching 87.27% wi...

  56. ParseFixer: An Agentic Framework for Document Parsing via Selective Multimodal Correction

    cs.CV 2026-06 unverdicted novelty 3.0

    ParseFixer combines full-page backbone parsing with agentic selective multimodal correction to reach third place (score 61.78) in the DataMFM Challenge Track 1.

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

    cs.CV 2026-06 unverdicted novelty 3.0

    PaddleOCR-VL-1.6 improves on PaddleOCR-VL-1.5 via region-aware data optimization and progressive post-training to reach 96.33% on OmniDocBench v1.6.

Reference graph

Works this paper leans on

73 extracted references · 73 canonical work pages · cited by 50 Pith papers · 9 internal anchors

  1. [2]

    R. AI. Rolmocr: A faster, lighter open source ocr model, 2025

    work page 2025

  2. [3]

    Ernie 4.5 technical report, 2025

    Baidu-ERNIE-Team. Ernie 4.5 technical report, 2025

    work page 2025

  3. [4]

    Blecher, G

    L. Blecher, G. Cucurull, T. Scialom, and R. Stojnic. Nougat: Neural optical understanding for academic documents, 2023

    work page 2023

  4. [5]

    Pix2text

    breezedeus. Pix2text. https://github.com/breezedeus/Pix2Text, 2022. Accessed: 2025-06-23

    work page 2022

  5. [6]

    Casey and E

    R. Casey and E. Lecolinet. A survey of methods and strategies in character segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 18 0 (7): 0 690--706, 1996. doi:10.1109/34.506792

    work page doi:10.1109/34.506792 1996

  6. [8]

    C. Cui, T. Gao, S. Wei, Y. Du, R. Guo, S. Dong, B. Lu, Y. Zhou, X. Lv, Q. Liu, X. Hu, D. Yu, and Y. Ma. Pp-lcnet: A lightweight cpu convolutional neural network, 2021. URL https://arxiv.org/abs/2109.15099

    work page arXiv 2021

  7. [9]

    Docling Team . Docling . https://github.com/docling-project/docling, 2024. Accessed: 2025-06-23

    work page 2024

  8. [13]

    open-parse

    Filimoa. open-parse. https://github.com/Filimoa/open-parse, 2024. Accessed: 2025-06-23

    work page 2024

  9. [14]

    I. J. Goodfellow, Y. Bulatov, J. Ibarz, S. Arnoud, and V. Shet. Multi-digit number recognition from street view imagery using deep convolutional neural networks, 2014. URL https://arxiv.org/abs/1312.6082

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  10. [16]

    J. Ha, R. M. Haralick, and I. T. Phillips. Recursive xy cut using bounding boxes of connected components. In Proceedings of 3rd International Conference on Document Analysis and Recognition, volume 2, pages 952--955. IEEE, 1995

    work page 1995

  11. [17]

    hiroi sora. Umi-ocr. https://github.com/hiroi-sora/Umi-OCR, 2022. Accessed: 2025-06-23

    work page 2022

  12. [18]

    W. Hu, X. Cai, J. Hou, S. Yi, and Z. Lin. Gtc: Guided training of ctc towards efficient and accurate scene text recognition. In Proceedings of the AAAI conference on artificial intelligence, volume 34, pages 11005--11012, 2020

    work page 2020

  13. [19]

    OpenVINO Toolkit

    Intel Corporation . OpenVINO Toolkit . https://www.intel.com/content/www/us/en/developer/tools/openvino-toolkit/overview.html, 2018. Accessed: 2025-06-23

    work page 2018

  14. [20]

    KevinHuSh. ragflow. https://github.com/infiniflow/ragflow, 2023. Accessed: 2025-06-23

    work page 2023

  15. [21]

    u ttler, M. Lewis, W.-t. Yih, T. Rockt \

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V. Karpukhin, N. Goyal, H. K \"u ttler, M. Lewis, W.-t. Yih, T. Rockt \"a schel, et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems, 33: 0 9459--9474, 2020

    work page 2020

  16. [26]

    Y. Ma, D. Yu, T. Wu, and H. Wang. Paddlepaddle: An open-source deep learning platform from industrial practice. Frontiers of Data and Domputing, 1 0 (1): 0 105--115, 2019

    work page 2019

  17. [27]

    ONNX Runtime

    Microsoft Corporation . ONNX Runtime . https://github.com/microsoft/onnxruntime, 2018. Accessed: 2025-06-23

    work page 2018

  18. [28]

    S. Mori, H. Nishida, and H. Yamada. Optical Character Recognition. John Wiley & Sons, 1999

    work page 1999

  19. [31]

    TensorRT

    NVIDIA Corporation . TensorRT . https://developer.nvidia.com/tensorrt, 2017. Accessed: 2025-06-23

    work page 2017

  20. [32]

    Triton Inference Server

    NVIDIA Corporation . Triton Inference Server . https://github.com/triton-inference-server/server, 2018. Accessed: 2025-06-23

    work page 2018

  21. [33]

    Ouyang, Y

    L. Ouyang, Y. Qu, H. Zhou, J. Zhu, R. Zhang, Q. Lin, B. Wang, Z. Zhao, M. Jiang, X. Zhao, et al. Omnidocbench: Benchmarking diverse pdf document parsing with comprehensive annotations. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 24838--24848, 2025

    work page 2025

  22. [34]

    Ai studio

    PaddlePaddle Team . Ai studio. https://aistudio.baidu.com, 2019. Accessed: 2025-06-23

    work page 2019

  23. [35]

    Paruchuri

    V. Paruchuri. Marker. https://github.com/VikParuchuri/marker, 2023. Accessed: 2025-06-23

    work page 2023

  24. [37]

    S. Ramírez. FastAPI . https://github.com/fastapi/fastapi, 2018. Accessed: 2025-06-23

    work page 2018

  25. [38]

    B. Shi, X. Bai, and C. Yao. An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition, 2015. URL https://arxiv.org/abs/1507.05717

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  26. [40]

    unstructured

    Unstructured-IO. unstructured. https://github.com/Unstructured-IO/unstructured, 2022. Accessed: 2025-06-23

    work page 2022

  27. [41]

    Verhoeven, T

    F. Verhoeven, T. Magne, and O. Sorkine-Hornung. Uvdoc: neural grid-based document unwarping. In SIGGRAPH Asia 2023 Conference Papers, pages 1--11, 2023

    work page 2023

  28. [43]

    H. Wei, C. Liu, J. Chen, J. Wang, L. Kong, Y. Xu, Z. Ge, L. Zhao, J. Sun, Y. Peng, et al. General ocr theory: Towards ocr-2.0 via a unified end-to-end model. 2024

    work page 2024

  29. [46]

    2021 , eprint=

    PP-LCNet: A Lightweight CPU Convolutional Neural Network , author=. 2021 , eprint=

    work page 2021

  30. [47]

    General ocr theory: Towards ocr-2.0 via a unified end-to-end model , author=

    work page

  31. [48]

    Distilling the Knowledge in a Neural Network

    Distilling the knowledge in a neural network , author=. arXiv preprint arXiv:1503.02531 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  32. [49]

    Pp-formulanet: Bridging accuracy and efficiency in advanced formula recognition.arXiv preprint arXiv:2503.18382, 2025

    PP-FormulaNet: Bridging Accuracy and Efficiency in Advanced Formula Recognition , author=. arXiv preprint arXiv:2503.18382 , year=

    work page arXiv

  33. [50]

    Proceedings of 3rd International Conference on Document Analysis and Recognition , volume=

    Recursive XY cut using bounding boxes of connected components , author=. Proceedings of 3rd International Conference on Document Analysis and Recognition , volume=. 1995 , organization=

    work page 1995

  34. [51]

    Pp-ocrv3: More attempts for the improvement of ultra lightweight ocr system

    PP-OCRv3: More attempts for the improvement of ultra lightweight OCR system , author=. arXiv preprint arXiv:2206.03001 , year=

    work page arXiv

  35. [52]

    arXiv preprint arXiv:2503.04065 , year=

    PP-DocBee: Improving Multimodal Document Understanding Through a Bag of Tricks , author=. arXiv preprint arXiv:2503.04065 , year=

    work page arXiv

  36. [53]

    2025 , eprint=

    ERNIE 4.5 Technical Report , author=. 2025 , eprint=

    work page 2025

  37. [54]

    arXiv preprint arXiv:2210.05391 , year=

    Pp-structurev2: A stronger document analysis system , author=. arXiv preprint arXiv:2210.05391 , year=

    work page arXiv

  38. [55]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Gtc: Guided training of ctc towards efficient and accurate scene text recognition , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

    work page

  39. [56]

    arXiv preprint arXiv:2205.00159 , year=

    Svtr: Scene text recognition with a single visual model , author=. arXiv preprint arXiv:2205.00159 , year=

    work page arXiv

  40. [57]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Omnidocbench: Benchmarking diverse pdf document parsing with comprehensive annotations , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    work page

  41. [58]

    MinerU: An Open-Source Solution for Precise Document Content Extraction

    Mineru: An open-source solution for precise document content extraction , author=. arXiv preprint arXiv:2409.18839 , year=

    work page internal anchor Pith review arXiv

  42. [59]

    2023 , howpublished =

    Vik Paruchuri , title=. 2023 , howpublished =

    work page 2023

  43. [60]

    2022 , howpublished =

    breezedeus , title=. 2022 , howpublished =

    work page 2022

  44. [61]

    2022 , howpublished =

    Unstructured-IO , title=. 2022 , howpublished =

    work page 2022

  45. [62]

    2024 , howpublished =

    Filimoa , title=. 2024 , howpublished =

    work page 2024

  46. [63]

    2024 , howpublished =

    work page 2024

  47. [64]

    Poznanski, A

    olmocr: Unlocking trillions of tokens in pdfs with vision language models , author=. arXiv preprint arXiv:2502.18443 , year=

    work page arXiv

  48. [65]

    arXiv preprint arXiv:2503.11576 (2025)

    SmolDocling: An ultra-compact vision-language model for end-to-end multi-modal document conversion , author=. arXiv preprint arXiv:2503.11576 , year=

    work page arXiv

  49. [66]

    2023 , eprint=

    Nougat: Neural Optical Understanding for Academic Documents , author=. 2023 , eprint=

    work page 2023

  50. [67]

    Qwen2.5 Technical Report

    Qwen2.5 Technical Report , author =. arXiv preprint arXiv:2412.15115 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  51. [68]

    Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling

    Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling , author=. arXiv preprint arXiv:2412.05271 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  52. [69]

    2018 , howpublished =

    work page 2018

  53. [70]

    2017 , howpublished =

    work page 2017

  54. [71]

    2018 , howpublished =

    Sebastián Ramírez , title =. 2018 , howpublished =

    work page 2018

  55. [72]

    2019 , howpublished =

    AI Studio , author =. 2019 , howpublished =

    work page 2019

  56. [73]

    SIGGRAPH Asia 2023 Conference Papers , pages=

    UVDoc: neural grid-based document unwarping , author=. SIGGRAPH Asia 2023 Conference Papers , pages=

    work page 2023

  57. [74]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Real-time scene text detection with differentiable binarization , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

    work page

  58. [75]

    Reducto AI , title =

    work page

  59. [76]

    Pp-doclayout: A unified document layout detection model to accelerate large-scale data construction.arXiv preprint arXiv:2503.17213, 2025

    PP-DocLayout: A Unified Document Layout Detection Model to Accelerate Large-Scale Data Construction , author=. arXiv preprint arXiv:2503.17213 , year=

    work page arXiv

  60. [77]

    arXiv preprint arXiv:2107.02137 , year=

    Ernie 3.0: Large-scale knowledge enhanced pre-training for language understanding and generation , author=. arXiv preprint arXiv:2107.02137 , year=

    work page arXiv

  61. [78]

    GPT-4 Technical Report

    Gpt-4 technical report , author=. arXiv preprint arXiv:2303.08774 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  62. [79]

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning , author=. arXiv preprint arXiv:2501.12948 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  63. [80]

    Qwen3 Technical Report

    Qwen3 technical report , author=. arXiv preprint arXiv:2505.09388 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  64. [81]

    arXiv preprint arXiv:2009.09941 (2020)

    Pp-ocr: A practical ultra lightweight ocr system , author=. arXiv preprint arXiv:2009.09941 , year=

    work page arXiv 2009

  65. [82]

    Pp-ocrv2: Bag of tricks for ultra lightweight ocr system

    Pp-ocrv2: Bag of tricks for ultra lightweight ocr system , author=. arXiv preprint arXiv:2109.03144 , year=

    work page arXiv

  66. [83]

    2023 , howpublished =

    KevinHuSh , title=. 2023 , howpublished =

    work page 2023

  67. [84]

    2022 , howpublished =

    hiroi-sora , title=. 2022 , howpublished =

    work page 2022

  68. [85]

    Advances in neural information processing systems , volume=

    Retrieval-augmented generation for knowledge-intensive nlp tasks , author=. Advances in neural information processing systems , volume=

    work page

  69. [86]

    Frontiers of Data and Domputing , volume=

    PaddlePaddle: An open-source deep learning platform from industrial practice , author=. Frontiers of Data and Domputing , volume=

    work page

  70. [87]

    1999 , publisher=

    Optical Character Recognition , author=. 1999 , publisher=

    work page 1999

  71. [88]

    and Lecolinet, E

    Casey, R.G. and Lecolinet, E. , journal=. A survey of methods and strategies in character segmentation , year=

    work page

  72. [89]

    2014 , eprint=

    Multi-digit Number Recognition from Street View Imagery using Deep Convolutional Neural Networks , author=. 2014 , eprint=

    work page 2014

  73. [90]

    2015 , eprint=

    An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition , author=. 2015 , eprint=

    work page 2015