When it comes to the training dataset, the results of training YOLO, DocExtractor, and Faster-RCNN on both S-VED and S-VED e x t e n d e d prove to be similar, with an almost insignificant drop in AP when switching from the latter to the former.
← all excerpts
Cor<i>Deep</i> and the Sacrobosco Dataset: Detection of Visual Elements in Historical Documents.
1
—
—