dnn: moved NMSBoxes to imgproc module#23711
Conversation
|
IMO, it should be (but need to resolve dependency issue). |
|
Originally I wanted to move all NMS stuff to the core module to be close with It seems impossible to move NMS to objdetect because then dnn would depend on it and we already have inverse dependency (objdetect on_dnn_). I think, ideally all geometry-related classes and algorithms should be moved to the core or a separate module - large part of |
This reverts commit be33c15.
I agree with this. If the API move is being proceeding still, we can leave the legacy API calling the new one from other module, leave a warning for API deprecation and remove the legacy one completely in some point. |
|
Discussed on OpenCV core team meating. Found alternative solution. Closed. |
|
@asmorkalov, Once an alternative solution will be published, may I ask to refer it here? Just to be on touch. Thanks in advance! |
Motivation: non-maximum suppression can be used not only as postprocessing in DNN-based algorithms, so we move it to imgproc module together with other geometry- and contour-related algorithms (bounding rect, intersections, etc.).
Notes:
cv::dnnnamespace and keeping old functions in dnn module (hidden from user, header file not installed). Maybe it is not really necessary because dnn ABI is still considered experimental.dnnnamespace in the imgproc module to fix it...? 🤔)