Dolphin: Document Image Parsing via Heterogeneous Anchor Prompting Article Swipe

PDF

Hao Feng , Wei Shu , Xiang Fei , Wei Shi , Yingdong Han , Lei Liao , Jinghui Lu , Binghong Wu , Qi Liu , Chih‐Yu Lin , Jingqun Tang , Hao Liu , Can Huang ·

YOU? · · 2025 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2505.14059

Document image parsing is challenging due to its complexly intertwined elements such as text paragraphs, figures, formulas, and tables. Current approaches either assemble specialized expert models or directly generate page-level content autoregressively, facing integration overhead, efficiency bottlenecks, and layout structure degradation despite their decent performance. To address these limitations, we present \textit{Dolphin} (\textit{\textbf{Do}cument Image \textbf{P}arsing via \textbf{H}eterogeneous Anchor Prompt\textbf{in}g}), a novel multimodal document image parsing model following an analyze-then-parse paradigm. In the first stage, Dolphin generates a sequence of layout elements in reading order. These heterogeneous elements, serving as anchors and coupled with task-specific prompts, are fed back to Dolphin for parallel content parsing in the second stage. To train Dolphin, we construct a large-scale dataset of over 30 million samples, covering multi-granularity parsing tasks. Through comprehensive evaluations on both prevalent benchmarks and self-constructed ones, Dolphin achieves state-of-the-art performance across diverse page-level and element-level settings, while ensuring superior efficiency through its lightweight architecture and parallel parsing mechanism. The code and pre-trained models are publicly available at https://github.com/ByteDance/Dolphin

Related Topics

Product Requirements Document

Particle Image Velocimetry

Public Image Of Mother Teresa

Image Rectification

Concepts

No concepts available.

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2505.14059
PDF: https://arxiv.org/pdf/2505.14059
OA Status: green
OpenAlex ID: https://openalex.org/W4417298651

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4417298651

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2505.14059

Digital Object Identifier
Title: Dolphin: Document Image Parsing via Heterogeneous Anchor Prompting

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2025

Year of publication
Publication date: 2025-05-20

Full publication date if available
Authors: Hao Feng, Wei Shu, Xiang Fei, Wei Shi, Yingdong Han, Lei Liao, Jinghui Lu, Binghong Wu, Qi Liu, Chih‐Yu Lin, Jingqun Tang, Hao Liu, Can Huang

List of authors in order
Landing page: https://arxiv.org/abs/2505.14059

Publisher landing page
PDF URL: https://arxiv.org/pdf/2505.14059

Direct link to full text PDF
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://arxiv.org/pdf/2505.14059

Direct OA link when available
Cited by: 0

Total citation count in OpenAlex

Full payload

id	https://openalex.org/W4417298651
doi	https://doi.org/10.48550/arxiv.2505.14059
ids.doi	https://doi.org/10.48550/arxiv.2505.14059
ids.openalex	https://openalex.org/W4417298651
fwci
type	preprint
title	Dolphin: Document Image Parsing via Heterogeneous Anchor Prompting
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
is_xpac	False
apc_list
apc_paid
language	en
locations[0].id	pmh:oai:arXiv.org:2505.14059
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S4306400194
locations[0].source.issn
locations[0].source.type	repository
locations[0].source.is_oa	True
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	arXiv (Cornell University)
locations[0].source.host_organization	https://openalex.org/I205783295
locations[0].source.host_organization_name	Cornell University
locations[0].source.host_organization_lineage	https://openalex.org/I205783295
locations[0].license
locations[0].pdf_url	https://arxiv.org/pdf/2505.14059
locations[0].version	submittedVersion
locations[0].raw_type	text
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published	False
locations[0].raw_source_name
locations[0].landing_page_url	http://arxiv.org/abs/2505.14059
locations[1].id	doi:10.48550/arxiv.2505.14059
locations[1].is_oa	True
locations[1].source.id	https://openalex.org/S4306400194
locations[1].source.issn
locations[1].source.type	repository
locations[1].source.is_oa	True
locations[1].source.issn_l
locations[1].source.is_core	False
locations[1].source.is_in_doaj	False
locations[1].source.display_name	arXiv (Cornell University)
locations[1].source.host_organization	https://openalex.org/I205783295
locations[1].source.host_organization_name	Cornell University
locations[1].source.host_organization_lineage	https://openalex.org/I205783295
locations[1].license
locations[1].pdf_url
locations[1].version
locations[1].raw_type	article
locations[1].license_id
locations[1].is_accepted	False
locations[1].is_published
locations[1].raw_source_name
locations[1].landing_page_url	https://doi.org/10.48550/arxiv.2505.14059
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5005177314
authorships[0].author.orcid	https://orcid.org/0000-0002-0899-0438
authorships[0].author.display_name	Hao Feng
authorships[0].author_position	first
authorships[0].raw_author_name	Feng, Hao
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5030583530
authorships[1].author.orcid	https://orcid.org/0000-0003-0890-2634
authorships[1].author.display_name	Wei Shu
authorships[1].author_position	middle
authorships[1].raw_author_name	Wei, Shu
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5100660005
authorships[2].author.orcid	https://orcid.org/0000-0001-8271-1815
authorships[2].author.display_name	Xiang Fei
authorships[2].author_position	middle
authorships[2].raw_author_name	Fei, Xiang
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A5108434351
authorships[3].author.orcid	https://orcid.org/0000-0003-3554-0202
authorships[3].author.display_name	Wei Shi
authorships[3].author_position	middle
authorships[3].raw_author_name	Shi, Wei
authorships[3].is_corresponding	False
authorships[4].author.id	https://openalex.org/A5049863437
authorships[4].author.orcid	https://orcid.org/0000-0002-1093-1381
authorships[4].author.display_name	Yingdong Han
authorships[4].author_position	middle
authorships[4].raw_author_name	Han, Yingdong
authorships[4].is_corresponding	False
authorships[5].author.id	https://openalex.org/A5036022527
authorships[5].author.orcid	https://orcid.org/0000-0003-1325-2410
authorships[5].author.display_name	Lei Liao
authorships[5].author_position	middle
authorships[5].raw_author_name	Liao, Lei
authorships[5].is_corresponding	False
authorships[6].author.id	https://openalex.org/A5050212540
authorships[6].author.orcid	https://orcid.org/0000-0001-7149-6961
authorships[6].author.display_name	Jinghui Lu
authorships[6].author_position	middle
authorships[6].raw_author_name	Lu, Jinghui
authorships[6].is_corresponding	False
authorships[7].author.id	https://openalex.org/A5036758849
authorships[7].author.orcid	https://orcid.org/0000-0002-3361-2260
authorships[7].author.display_name	Binghong Wu
authorships[7].author_position	middle
authorships[7].raw_author_name	Wu, Binghong
authorships[7].is_corresponding	False
authorships[8].author.id	https://openalex.org/A5100393403
authorships[8].author.orcid	https://orcid.org/0009-0000-4144-938X
authorships[8].author.display_name	Qi Liu
authorships[8].author_position	middle
authorships[8].raw_author_name	Liu, Qi
authorships[8].is_corresponding	False
authorships[9].author.id	https://openalex.org/A5101652889
authorships[9].author.orcid	https://orcid.org/0000-0003-4661-4131
authorships[9].author.display_name	Chih‐Yu Lin
authorships[9].author_position	middle
authorships[9].raw_author_name	Lin, Chunhui
authorships[9].is_corresponding	False
authorships[10].author.id	https://openalex.org/A5100681719
authorships[10].author.orcid	https://orcid.org/0000-0003-2577-0119
authorships[10].author.display_name	Jingqun Tang
authorships[10].author_position	middle
authorships[10].raw_author_name	Tang, Jingqun
authorships[10].is_corresponding	False
authorships[11].author.id	https://openalex.org/A5100458818
authorships[11].author.orcid	https://orcid.org/0000-0002-5248-935X
authorships[11].author.display_name	Hao Liu
authorships[11].author_position	middle
authorships[11].raw_author_name	Liu, Hao
authorships[11].is_corresponding	False
authorships[12].author.id	https://openalex.org/A5035977616
authorships[12].author.orcid	https://orcid.org/0000-0001-7395-2270
authorships[12].author.display_name	Can Huang
authorships[12].author_position	last
authorships[12].raw_author_name	Huang, Can
authorships[12].is_corresponding	False
has_content.pdf	True
has_content.grobid_xml	True
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://arxiv.org/pdf/2505.14059
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-10-10T00:00:00
display_name	Dolphin: Document Image Parsing via Heterogeneous Anchor Prompting
has_fulltext	False
is_retracted	False
updated_date	2025-12-13T14:46:20.555162
primary_topic
cited_by_count	0
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2505.14059
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S4306400194
best_oa_location.source.issn
best_oa_location.source.type	repository
best_oa_location.source.is_oa	True
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	arXiv (Cornell University)
best_oa_location.source.host_organization	https://openalex.org/I205783295
best_oa_location.source.host_organization_name	Cornell University
best_oa_location.source.host_organization_lineage	https://openalex.org/I205783295
best_oa_location.license
best_oa_location.pdf_url	https://arxiv.org/pdf/2505.14059
best_oa_location.version	submittedVersion
best_oa_location.raw_type	text
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	http://arxiv.org/abs/2505.14059
primary_location.id	pmh:oai:arXiv.org:2505.14059
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S4306400194
primary_location.source.issn
primary_location.source.type	repository
primary_location.source.is_oa	True
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	arXiv (Cornell University)
primary_location.source.host_organization	https://openalex.org/I205783295
primary_location.source.host_organization_name	Cornell University
primary_location.source.host_organization_lineage	https://openalex.org/I205783295
primary_location.license
primary_location.pdf_url	https://arxiv.org/pdf/2505.14059
primary_location.version	submittedVersion
primary_location.raw_type	text
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	http://arxiv.org/abs/2505.14059
publication_date	2025-05-20
publication_year	2025
referenced_works_count	0
abstract_inverted_index.a	59, 76, 113
abstract_inverted_index.30	118
abstract_inverted_index.In	70
abstract_inverted_index.To	45, 108
abstract_inverted_index.an	67
abstract_inverted_index.as	12, 88
abstract_inverted_index.at	165
abstract_inverted_index.in	81, 104
abstract_inverted_index.is	3
abstract_inverted_index.of	78, 116
abstract_inverted_index.on	128
abstract_inverted_index.or	26
abstract_inverted_index.to	6, 98
abstract_inverted_index.we	49, 111
abstract_inverted_index.The	157
abstract_inverted_index.and	17, 37, 90, 132, 142, 153, 159
abstract_inverted_index.are	95, 162
abstract_inverted_index.due	5
abstract_inverted_index.fed	96
abstract_inverted_index.for	100
abstract_inverted_index.its	7, 150
abstract_inverted_index.the	71, 105
abstract_inverted_index.via	55
abstract_inverted_index.back	97
abstract_inverted_index.both	129
abstract_inverted_index.code	158
abstract_inverted_index.over	117
abstract_inverted_index.such	11
abstract_inverted_index.text	13
abstract_inverted_index.with	92
abstract_inverted_index.Image	53
abstract_inverted_index.These	84
abstract_inverted_index.first	72
abstract_inverted_index.image	1, 63
abstract_inverted_index.model	65
abstract_inverted_index.novel	60
abstract_inverted_index.ones,	134
abstract_inverted_index.their	42
abstract_inverted_index.these	47
abstract_inverted_index.train	109
abstract_inverted_index.while	145
abstract_inverted_index.Anchor	57
abstract_inverted_index.across	139
abstract_inverted_index.decent	43
abstract_inverted_index.either	21
abstract_inverted_index.expert	24
abstract_inverted_index.facing	32
abstract_inverted_index.layout	38, 79
abstract_inverted_index.models	25, 161
abstract_inverted_index.order.	83
abstract_inverted_index.second	106
abstract_inverted_index.stage,	73
abstract_inverted_index.stage.	107
abstract_inverted_index.tasks.	124
abstract_inverted_index.Current	19
abstract_inverted_index.Dolphin	74, 99, 135
abstract_inverted_index.Through	125
abstract_inverted_index.address	46
abstract_inverted_index.anchors	89
abstract_inverted_index.content	30, 102
abstract_inverted_index.coupled	91
abstract_inverted_index.dataset	115
abstract_inverted_index.despite	41
abstract_inverted_index.diverse	140
abstract_inverted_index.million	119
abstract_inverted_index.parsing	2, 64, 103, 123, 155
abstract_inverted_index.present	50
abstract_inverted_index.reading	82
abstract_inverted_index.serving	87
abstract_inverted_index.tables.	18
abstract_inverted_index.through	149
abstract_inverted_index.Document	0
abstract_inverted_index.Dolphin,	110
abstract_inverted_index.achieves	136
abstract_inverted_index.assemble	22
abstract_inverted_index.covering	121
abstract_inverted_index.directly	27
abstract_inverted_index.document	62
abstract_inverted_index.elements	10, 80
abstract_inverted_index.ensuring	146
abstract_inverted_index.figures,	15
abstract_inverted_index.generate	28
abstract_inverted_index.parallel	101, 154
abstract_inverted_index.prompts,	94
abstract_inverted_index.publicly	163
abstract_inverted_index.samples,	120
abstract_inverted_index.sequence	77
abstract_inverted_index.superior	147
abstract_inverted_index.available	164
abstract_inverted_index.complexly	8
abstract_inverted_index.construct	112
abstract_inverted_index.elements,	86
abstract_inverted_index.following	66
abstract_inverted_index.formulas,	16
abstract_inverted_index.generates	75
abstract_inverted_index.overhead,	34
abstract_inverted_index.paradigm.	69
abstract_inverted_index.prevalent	130
abstract_inverted_index.settings,	144
abstract_inverted_index.structure	39
abstract_inverted_index.approaches	20
abstract_inverted_index.benchmarks	131
abstract_inverted_index.efficiency	35, 148
abstract_inverted_index.mechanism.	156
abstract_inverted_index.multimodal	61
abstract_inverted_index.page-level	29, 141
abstract_inverted_index.challenging	4
abstract_inverted_index.degradation	40
abstract_inverted_index.evaluations	127
abstract_inverted_index.integration	33
abstract_inverted_index.intertwined	9
abstract_inverted_index.large-scale	114
abstract_inverted_index.lightweight	151
abstract_inverted_index.paragraphs,	14
abstract_inverted_index.performance	138
abstract_inverted_index.pre-trained	160
abstract_inverted_index.specialized	23
abstract_inverted_index.architecture	152
abstract_inverted_index.bottlenecks,	36
abstract_inverted_index.limitations,	48
abstract_inverted_index.performance.	44
abstract_inverted_index.comprehensive	126
abstract_inverted_index.element-level	143
abstract_inverted_index.heterogeneous	85
abstract_inverted_index.task-specific	93
abstract_inverted_index.\textbf{P}arsing	54
abstract_inverted_index.\textit{Dolphin}	51
abstract_inverted_index.self-constructed	133
abstract_inverted_index.state-of-the-art	137
abstract_inverted_index.autoregressively,	31
abstract_inverted_index.multi-granularity	122
abstract_inverted_index.analyze-then-parse	68
abstract_inverted_index.Prompt\textbf{in}g}),	58
abstract_inverted_index.\textbf{H}eterogeneous	56
abstract_inverted_index.(\textit{\textbf{Do}cument	52
abstract_inverted_index.https://github.com/ByteDance/Dolphin	166
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	13
citation_normalized_percentile