Possible to train with own images about ocr

[mateogianolio]

If you know the font in advance I would say the easiest way is to put that font in the config file (and tweak size, etc to your liking) and let the network train itself. See 'Configuration' and 'Usage' in the README.

Then you can use the trained network on your images, but you have to first convert them to a bit array of the size specified in the config file (see the functions parse and mnist in main.js for conversion examples).

Either way you probably have to write your own parsing function.

[potmo]

Great thanks. I'll try that out