ConceptNon-Maximum Suppression / 非极大值抑制

NMS

A commonly used post-processing algorithm in object detection that retains the best results by suppressing redundant detection boxes with high overlap, typically measured using IoU as the overlap metric.

Timeline (last 90 days)

Oct 13

In the merge step of tiled inference, IoU-based NMS alone is insufficient; intersection-over-smaller (IoS) needs to be added as a merge criterion.

Unverified50%

Sep 4

DETR abandons the anchor box design and non-maximum suppression (NMS) post-processing pipeline found in traditional detectors.

Unverified50%

All Facts (2)

Unverified

In the merge step of tiled inference, IoU-based NMS alone is insufficient; intersection-over-smaller (IoS) needs to be added as a merge criterion.

50%Unverified

DETR abandons the anchor box design and non-maximum suppression (NMS) post-processing pipeline found in traditional detectors.

50%

Source Articles

A Beginner's Guide to Computer Vision: From Pixels to YOLO and Image SegmentationOct 13 The CoreML Model Is Fine, Your Photos Are Too Big: The Image Pipeline Trap in On-Device AIOct 13