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.
This PR sets the minimal area required to return a box in the RTree intersection routine to 0. The initial value of 1. was based on the misconception that all bboxes' coordinates were expressed in pixels, hence no intersection could be smaller than 1 (pixel).
Users should be able to express their bbox in real world coordinates, or even normalize them, hence have area much smaller than 1., yet still having an IoU with the bbox in reference higher than the threshold.
This PR sets the minimal area required to return a box in the
RTreeintersection routine to0.The initial value of1.was based on the misconception that all bboxes' coordinates were expressed in pixels, hence no intersection could be smaller than 1 (pixel).Users should be able to express their bbox in real world coordinates, or even normalize them, hence have area much smaller than 1., yet still having an IoU with the bbox in reference higher than the threshold.
Closes #29