Bbox coordinates

[vetwl]

Hello, i am struggling with the problem of obtaining coordinates as an output of the network. I do not fully understand the meaning of an output. I am getting "bbox_pred" values approximately in range of -1 - 1, sometimes 0,001, sometimes bigger then 1, sometimes lesser then -1, is it ok? As i think: "bbox_pred" is for sorta "deltas" for bounding box, that is the result of regression and the bbox coordinates should somehow be obtained from rois+bbox_pred, am i right?. I was looking for answer through issues and some other resources but have not found it helpful, may be i am just misunderstood something. Could u please describe how the coordinates are obtained? Not clipping anything, but if u want to help, i will send my config/proto/code. I would really appreciate this, as i am got stuck in a while. Sorry for any grammar mistakes i could have made.

[vetwl]

I was lucky to manage the problem myself, i was partly right about how predict boxes should be built, there is a function for testing, that works, like i have said, with proposal areas and output of the network for the deltas. For now i am trying to obtain decent detection result that are not even good yet. I asked same question on reddit and one guy propose his solution, that didn't work for me, but mb u will find it useful (reddit: https://www.reddit.com/r/learnpython/comments/fz39ss/faster_rcnn/)