What files go into Training_Data folder, and Validation_Data?

[ciobania]

Hello,

Thanks for sharing this, and making it easier to work with.

What exactly goes into the .box files? Are those files generated via JTextEditor/QtTextEditor, per line, or per character?

How come we don't need truth files, like the do umentation says - albeit it has not changed in years, so it may be obsolete.

[RawthiL]

Hello,

The .box files contain whole words, but separated in letters, something like this:

H 237 2686 593 2743 0
E 237 2686 593 2743 0
L 237 2686 593 2743 0
L 237 2686 593 2743 0
O 237 2686 593 2743 0
	 237 2686 593 2743 0
T 242 2625 735 2676 0
H 242 2625 735 2676 0
E 242 2625 735 2676 0
R 242 2625 735 2676 0
E 242 2625 735 2676 0

Those are two separate words, in two different bounding boxes (BBoxes). Each letter has its own entry, but the BBox is the same for the whole words.

I build this .box files using the annotated file from VIA software. The VIA software creates a .json file which I parsed to generate the .box files and the .tiff images.
I tried to use the JTextEditor/QtTextEditor software, but I needed some extra functionality that they did not have (nothing to do with the OCR training). I have no experience ussing them.

Regarding the truth files, I'm not sure which files you mean. I trained the Tesseract 5.x LSTM models, they only need the .box and .tiff to create the .lstm files which are used for training.