MIT Press

Table 1:

Details for the three datasets in the S2-VLUE benchmark.

	GROTOAP2	DocBank	S2-VL
Train / Dev / Test Pages	83k / 18k / 18k	398k / 50k / 50k	1.3k^¹
Annotation Method	Automatic	Automatic	Human Annotation
Scientific Discipline	Life Science	Math / Physics / CS	19 Disciplines
Visual Layout Group	PDF parsing	Vision model	Gold Label / Detection methods
Number of Categories	22	12	15

Average Token Count^²	1203 (591)	838 (503)	790 (453)
Average Text Line Count	90 (51)	60 (34)	64 (54)
Average Text Block Count	12 (16)	15 (8)	22 (36)

	GROTOAP2	DocBank	S2-VL
Train / Dev / Test Pages	83k / 18k / 18k	398k / 50k / 50k	1.3k^¹
Annotation Method	Automatic	Automatic	Human Annotation
Scientific Discipline	Life Science	Math / Physics / CS	19 Disciplines
Visual Layout Group	PDF parsing	Vision model	Gold Label / Detection methods
Number of Categories	22	12	15

Average Token Count^²	1203 (591)	838 (503)	790 (453)
Average Text Line Count	90 (51)	60 (34)	64 (54)
Average Text Block Count	12 (16)	15 (8)	22 (36)

This is the total number of pages in the S2-VL dataset; we use 5-fold cross-validation for training and testing.

We report the average token, text line, and text block count per page, with standard deviations in parentheses.