Can you run EasyOCR using only text detection without running text recognition (dbnet18)

[R4Ajeti]

Hi, Thanks for this awesome project but why cannot find a way of running separately only text detection?

FIRST USE CASE: I need to run this using only the CPU. And I have multiple folders. Each folder contains a huge amount of images like 900+ images per folder where the image resolution is 1280x400. And I wanna run as fast as possible I have heard that "dbnet18" is the fastest model. Can someone give me a hint or example code on how to archive this using this package?!

SECOND USE CASE: I should train any existing model to my image dataset. Any suggestion about the best model that is super fast and not so bad in terms of accuracy? And also how do I perform training on this package?

PS: I am coming from paddleOCR btw because they have a lot of bugs these days on macOS so time to change.

[daniellovera]

You can run just text detection. See the API documentation here: https://www.jaided.ai/easyocr/documentation/