What is Non-Maximum Suppression (NMS) and its variants?
mediumAnswer
- Post-processing that removes duplicate detections: sort boxes by confidence, keep the top one, remove all others with IoU > threshold with it, repeat.
- Standard threshold: 0.5.
- Variants: Soft-NMS (lower the score of overlapping boxes instead of removing, better for occluded objects), matrix NMS (used in SOLOv2 segmentation), and class-aware vs class-agnostic.
- Recent detectors (DETR) avoid NMS entirely via set prediction.
Check yourself — multiple choice
- NMS creates duplicates
- Remove overlapping duplicate boxes by IoU threshold — variants: Soft-NMS, matrix NMS; DETR skips NMS
- NMS is a loss function
- Same as pooling
NMS: dedupe boxes by IoU; Soft-NMS penalizes; DETR eliminates the step.
#detection#computer-vision
Practise Deep Learning
214 interview questions in this topic.
Related questions
- How does a two-stage detector (Faster R-CNN) work?
- How does YOLO / SSD / RetinaNet differ from two-stage detection?
- What are anchor boxes and their downsides?
- How does Mask R-CNN extend Faster R-CNN for instance segmentation?
- What is a Feature Pyramid Network (FPN)?
- How does DETR reformulate object detection?