horizontal text dection

[daniiki]

Hi,
I know similar questions were already asked, but I could not solve my problem:
I have only horizontal bounding boxes, but the results after training are always rotated bounding boxes :(
I dont know if the error is related to training or to evalution. For evaluation I use the eval.py file.

• Is it possible to train without rotating the bounding boxes?
• why there are only the two cases angle>0 and angle<0 represented in the restore_rectangle_rbox function? I could image that this ist the problem, but I dont know how to solve this??