The problem of "-ocr_feed_list" config

MichalBusta / E2E-MLT

E2E-MLT - an Unconstrained End-to-End Method for Multi-Language Scene Text

MIT License

291 stars 84 forks source link

The problem of "-ocr_feed_list" config #65

Closed duxiangcheng closed 4 years ago

duxiangcheng commented 4 years ago

In the train configurations, I noticed the option "-ocr_feed_list". What does this option do?

duxiangcheng commented 4 years ago

@MichalBusta Please help me! thanks

MichalBusta commented 4 years ago

pointer to GT list of cropped words.

see: https://github.com/MichalBusta/E2E-MLT/tree/master/sample_train_data/MLT_CROPS

duxiangcheng commented 4 years ago

@MichalBusta But the data in the repo are images along with their gt. Where can I get the cropped images? During your training, what's the ratio of the two(-train_list and -ocr_feed_list)?

MichalBusta commented 4 years ago

Simply cut rotated rectangles from full GT images. additionally you can generate some synthetic data or use some already generated (https://www.robots.ox.ac.uk/~vgg/research/text/)