Page Filtering

Model to identify pages in children's books of the long 19th century (ca. 1789-1914) that contain illustrations. It is used to filter non-relevant pages without illustrations trained on hand-coded data.

Results on our validation dataset:

	f1score	precision	recall	accuracy
not-relevant	99.63	100	99.26	-
relevant-cover	85.71	75	100	-
relevant-page	100	100	100	-
Macro Avg.	95.11	91.67	99.75	99.37

Dataset:

	data	train	test
not-relevant	902	631	271
relevant-cover	20	14	6
relevant-page	136	95	41