Closed Zico2017 closed 5 years ago
Thanks for writing the issue.
We are going to improve editing the rotated box and the format.
Could you tell us which object detection tool are you going to use with the format? Because we want to know which format would be the best for the rotated box to connect to other tools.
Best regards, Ryo
A very prompt reply!
I only find two papers which use the format based on the code of faster r-cnn, it looks 5 variables to label a rotated bounding box is better than 8 variables. Arbitrary-Oriented Scene Text Detection via Rotation Proposals (https://github.com/mjq11302010044/RRPN) Learning a Rotation Invariant Detector with Rotatable Bounding Box (https://github.com/liulei01/DRBox)
Looking for your next version!
Thanks for letting us know those tools.
We could find the proposed format in "Arbitrary-Oriented Scene Text Detection via Rotation Proposals ".
We would let you know when we submit the new update.
Best regards, Ryo
The new update version 1.81 was released.
Let us know your opinion.
@ryouchinsa could you please share the code rotated_box
how to convert to polygon
I was annotate samples use rotated_box
The model I pick use ICDAR2015
format like 377,117,463,117,465,130,378,130,Genaxis Theatre
Hi,
This is how you could convert the rotated box to the polygon points.
cos = cos(rot);
sin = sin(rot);
x1 = -width / 2;
y1 = -height / 2;
x1 = cx + x1 * cos - y1 * sin;
y1 = cy + x1 * sin + y1 * cos;
x2 = width / 2;
y2 = -height / 2;
...
Anyway, your tool is the most convenient to use for OSX users, thank you!