Training on my own images

Question

Training on my own images

Opened this issue 6 years ago · 2 comments

@gaxler Thank you for your hard work.

How can I train your tool using my own images?
What is required, and what data format can I use? PAGE XML annotation? YOLO darknet bbox?

Answer 1 · 2018-09-08T00:52:05.000Z

You can write an interface to your own data. Basically you need to have a generator that produces MetaImage objects ( see data/metaimage.py). You can find examples in the data folder ( e.g. data/iclef.py)

Answer 2 · 2018-09-09T23:32:08.000Z

@gaxler how do your tool compare to ocroseg from ocropus3?
What label format does your tool accept: yolo, voc or kitti?