We present the Arxiv Figures & Tables Database (AFTdb), which consists of an aggregation of figures and tables from scientific articles sourced from the arXiv platform.

The purpose of this dataset is to train multimodal models specialized in images of document-type objects (graphs, functional diagrams, tables, etc.), rather than photographic-type images. The idea is that a model trained on this type of data will be more coherent within the context of document corpora than a model trained on pictorial compositions. To establish a connection between the two modalities (image and text), captions for each object are also provided. As captions can sometimes be very brief, the article's summary is also included to add context to the document object if necessary. All textual data (titles, abstracts, and captions) are available in both English (original language) and French through translation using Google Translate.

For this reason, a corpus of scientific articles was prioritized. Due to the scientific rigor demanded, each document-type object is systematically accompanied by a caption (similar to captions for pictorial images on platforms like Flickr, for example).

The database is divided into two types of document objects: figures and tables. For the table part, it is possible to approach two different types of learning. The first, similar to figures, associates the image with the caption. However, in the data field, the LaTeX source code of the table is also provided. An objective can be to take an image of a table and convert it into text using this source code.

Loading the database

The figure part is relatively substantial, and it is advisable to use the dataset in streaming mode:

aftdb_figure = load_dataset("cmarkea/aftdb", "figure", streaming=True)

The table part is less substantial and can be downloaded locally directly:

aftdb_table = load_dataset("cmarkea/aftdb", "table")

Both categories are compatible, and it is possible to load both types simultaneously:

aftdb = load_dataset("cmarkea/aftdb", "figure+table", streaming=True)

This is the default configuration.

Statistical Description

The descended articles correspond to a portion of the articles that had their last modifications in the year 2023 on the arXiv platform.

Number of
articles	22,893
authors	90,165
figures (train)	157,944
figures (test)	3,579
tables (train)	16,415
tables (test)	395
total words in English titles	234,072
total words in French titles	308,187
total words in English abstracts	3,879,940
total words in French abstracts	4,536,101
total words in English captions	7,689,270
total words in French captions	8,513,199

Here is the distribution of articles in the dataset by arXiv category.

categorie	Freq (%)	categorie	Freq (%)
cs.LG	7.29594	cs.AI	3.88624
cs.CV	2.48066	hep-ph	2.12586
astro-ph.SR	2.01854	astro-ph.GA	1.85782
stat.ME	1.77373	physics.flu-dyn	1.71847
cond-mat.stat-mech	1.66027	stat.ML	1.64265
eess.SP	1.63971	cs.CL	1.4838
astro-ph.HE	1.48087	hep-ex	1.43361
astro-ph.IM	1.43014	physics.comp-ph	1.39464
nucl-th	1.3925	math.NA	1.36794
hep-th	1.30467	physics.optics	1.28037
astro-ph.EP	1.19494	cond-mat.mtrl-sci	1.18373
cs.SY	1.17305	eess.SY	1.16131
stat.AP	1.14369	cs.IT	1.14022
math.IT	1.14022	physics.ins-det	1.1258
gr-qc	1.10845	cs.RO	1.10765
cond-mat.soft	1.05425	cond-mat.mes-hall	1.04277
astro-ph.CO	1.03743	math.OC	1.01047
cs.CR	0.994986	cond-mat.str-el	0.984041
cs.DC	0.972294	physics.chem-ph	0.95681
cond-mat.dis-nn	0.947199	cs.NI	0.941593
cond-mat.quant-gas	0.880191	physics.atom-ph	0.878322
cs.CE	0.874851	hep-lat	0.837476
cs.NE	0.836141	cs.SI	0.830001
math.DS	0.821992	eess.AS	0.813716
nucl-ex	0.810512	math-ph	0.808376
cs.HC	0.784616	cs.MM	0.709065
physics.app-ph	0.695182	cs.SD	0.694915
physics.plasm-ph	0.694381	cs.MA	0.693847
math.ST	0.682101	quant-ph	2.53645
stat.TH	0.682101	physics.bio-ph	0.650332
eess.IV	0.650065	physics.soc-ph	0.649531
cs.GR	0.633513	cs.IR	0.620965
cs.DB	0.620165	cs.CY	0.596404
cs.AR	0.576115	math.GT	0.555025
q-bio.QM	0.545948	physics.data-an	0.543812
math.CO	0.535269	math.PR	0.51845
physics.ao-ph	0.515246	nlin.CD	0.496559
stat.CO	0.49202	q-bio.PE	0.474934
cond-mat.supr-con	0.454378	q-bio.NC	0.453577
cs.GT	0.445301	econ.GN	0.429283
cs.SE	0.423143	econ.GN	0.429283
cs.ET	0.419405	physics.space-ph	0.394577
nlin.PS	0.368949	cs.PF	0.345188
physics.acc-ph	0.335845	cond-mat.other	0.331573
econ.EM	0.328903	physics.med-ph	0.320361
cs.DM	0.304876	math.AP	0.294198
nlin.AO	0.256555	q-bio.BM	0.235198
q-fin.CP	0.223184	math.AT	0.198624
cs.PL	0.192483	physics.class-ph	0.18661
math.DG	0.184741	q-fin.ST	0.181538
cs.LO	0.17433	cs.CC	0.153506
cs.DL	0.143895	q-fin.TR	0.136954
math.MG	0.135352	math.AG	0.134818
q-fin.MF	0.131615	q-bio.TO	0.126809
q-bio.GN	0.120936	math.SG	0.118266
math.GR	0.116665	math.CA	0.116398
math.CV	0.116398	cs.MS	0.110524
math.HO	0.106253	nlin.SI	0.104918
math.RT	0.100113	cs.FL	0.0995787
q-fin.PM	0.097176	econ.TH	0.0955742
math.SP	0.0880991	q-fin.GN	0.0875652
q-fin.RM	0.0859634	physics.ed-ph	0.0819589
math.QA	0.0787553	q-bio.CB	0.0752847
nlin.CG	0.072882	physics.atm-clus	0.072615
math.NT	0.0720811	math.FA	0.0712802
q-bio.MN	0.0707463	physics.pop-ph	0.064873
q-fin.PR	0.0635382	stat.OT	0.0619364
cs.OS	0.0544613	cs.SC	0.0467192
physics.gen-ph	0.0461853	physics.hist-ph	0.0429817
math.AC	0.0379093	q-bio.SC	0.0331039
math.CT	0.0309682	math.RA	0.0304342
math.GN	0.0274976	math.LO	0.0261628
cs.OH	0.0248279	math.GM	0.0168189
math.OA	0.016552	cs.GL	0.0114796
math.KT	0.00694114	q-bio.OT	0.00186877

Field Descriptions

id: Unique identifier for each observation.
paper_id: Unique arXiv identifier for each article.
type: 'figure' for graphic objects such as graphs, functional diagrams, etc., and 'table' for tables.
authors: Names of the article's authors.
categories: arXiv categories of the article.
title: Title of the article.
summary: Article summary.
caption: Caption of the document-type object.
image: Pillow image of the document-type object.
data: For figures, it represents the filename of the figure; for tables, it is the LaTeX transcription of the table.
newcommands: List containing the LaTeX newcommands used in the article.

Citation

@online{DeAFTdb,
  AUTHOR = {Cyrile Delestre},
  URL = {https://huggingface.co/datasets/cmarkea/aftdb},
  YEAR = {2024},
  KEYWORDS = {NLP ; Multimodal}
}