Large Scale Non maximum Suppression. This project implements a O(nlog(n)) approach for non max suppression, useful for object detection ran on very large images such as satellite or histology, when the number of instances to prune becomes very large.
As of version 0.3.1, rotated bboxes are not supported. This should be, considering that most object detection frameworks enable it.
This means:
Find a way for the lsnms.nms function to be fed with rotated bboxes (be given the four corners of the bbox for instance), but also accept standard (ie non rotated) bboxes
Implement rotated bboxes iou computation
Change the RTree building process (especially the node splitting part, and the box_englobing_boxes func)
As of version 0.3.1, rotated bboxes are not supported. This should be, considering that most object detection frameworks enable it.
This means:
lsnms.nmsfunction to be fed with rotated bboxes (be given the four corners of the bbox for instance), but also accept standard (ie non rotated) bboxesbox_englobing_boxesfunc)