Wrong clean region

[marcussacana]

First, thanks for the english optimizations, I hope this tool continue becaming better.
Anyway, I found that the tool tried to redraw the wrong region to clean the text,

This happened to me when I used --size 3000

[zyddnys]

Thank you for finding the problem.
The current text mask generation refinement involves a lot of hand craft method designed for lower resolution images(e.g. dilation kernel size), which leads to such erroneous text mask.
I am currently training a new model with better initial text mask proposal(ViT model) and better refinement(using CRF), improvements are to be expected.
No timeline for new model release as of now.

[marcussacana]

closing for now, thx