Faster R-CNN: Non-maximum Suppression
Advanced
The problem, from first principles
Detectors produce many overlapping boxes for the same object. Non-maximum suppression keeps the best-scoring box and removes any other box that overlaps it too much, leaving one detection per object.
From theory to code
Implement nms(boxes, scores, thr), returning the indices of the kept boxes in score order.
Constraints
- Overlap is measured with IoU; suppress boxes with IoU above
thr.
Hints
Hint 1
Sort the indices by score. Walk through them, keeping each box that has IoU at most thr with every box already kept.
Keywords
research-paperscomputer-visiondetectionobject-detection
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.