East Text Detector "cuts" letter while detecting text.

[Mice948]

Hey, I've been trying to use The East Text Detector for a little project I have, but I'm facing errors such as the EAST module cuts off the characters causing later for the OCR module to not function well. I am following this tutorial And HERE you can see an example for an image I gave the module to handle. I have the latest trained module, and Python 3.6.8 Is there any way to maybe increase the rectangle?

[TuulaP]

You can just add some buffer to the endX , endY, which then also expands the box, e.g. in the tutorial

    endX = int(endX * rW)+2
    endY = int(endY * rH)+3

In the long run that is not a good solution as depending on case you might need different kind of expansion/narrowing. But for what I noticed the eval.py solution here seems to be more precise, although , there are some overlaps with areas which the linked tutorial does not seem to have (at least with my data).

[mailcorahul]

@Mice948 EAST sometimes doesn't detect texts perfectly which are too large(or zoomed in). You can try to add padding around those images to see if it gets detected nicely. You can also try the above logic to increase the box size. Or you can finetune EAST model on your set of images.