search

clovaai / deep-text-recognition-benchmark

Text recognition (optical character recognition) with deep learning methods, ICCV 2019
Apache License 2.0
3.75k stars 1.1k forks source link

How to apply CRAFT as preprocessing? #153

Open heromanba opened 4 years ago

heromanba commented 4 years ago

Hi, thanks for sharing this great work. I noticed that on ICDAR2019 ArT results ranking table, there is a saying "Before text recognition, we used the text detector called CRAFT as a preprocessing step.". I am wondering how can you apply CRAFT as preprocessing? Do you train these two models jointly? Or just use CRAFT pretrained model's inference as preprocessing?

HennyJie commented 4 years ago

Any idea how to combine these two repositories to make it end to end now?

heromanba commented 4 years ago

Not yet. I think end-to-end training is difficult. It might be possible to train these two models separately and cascade them together in inference. But another problem is how to train CRAFT on cropped text images.

HennyJie commented 4 years ago

Not yet. I think end-to-end training is difficult. It might be possible to train these two models separately and cascade them together in inference. But another problem is how to train CRAFT on cropped text images.

Thanks for your reply! By the way, I switched to another end-to-end repository https://github.com/MalongTech/research-charnet for my need. Check the end-to-end Text Recognition in this list https://github.com/hwalsuklee/awesome-deep-text-detection-recognition if you need it. : )

heromanba commented 4 years ago

Thanks, it's very helpful. ^_^

Amithvassit commented 4 years ago

Craft is available in pip .check

indykish commented 4 years ago

Not yet. I think end-to-end training is difficult. It might be possible to train these two models separately and cascade them together in inference. But another problem is how to train CRAFT on cropped text images.

Thanks for your reply! By the way, I switched to another end-to-end repository https://github.com/MalongTech/research-charnet for my need. Check the end-to-end Text Recognition in this list https://github.com/hwalsuklee/awesome-deep-text-detection-recognition if you need it. : )

The repo https://github.com/MalongTech/research-charnet doesn't produce the desired results.