oriented bounding boxes

[Arthur]

Hello !
Do you have any idea about what changes in the code are necessary to go from (xmax, xmin, ymin, ymax) labelling to (x1,y1,x2,y2,x3,y3,x4,y4) polynomial labelling to support oriented detection ? I already have an annoted dataset in (x1,y1,x2,y2,x3,y3,x4,y4) format converted to tfrecord but I do not know what to change in what function in the code in order to make it perform the training with the good loss function etc... Do you have any leads ?