SSD

Single Shot MultiBox Detector

SSD is a one-stage algorithm that first came up with Wei Liu in ECCV 2016.

It processes classification and bounding box regression at the same time as YOLO, but also adds a new idea, prior box, that is like anchor in Faster R-CNN.

The other difference is that instead of only selecting the top layer feature maps, SSD adds Pyramidal Feature Hierarchy to detect objects.

While selecting feature maps, input images will be separated into S x S small blocks. If one point P is selected as a center point, there will be several prior boxes around this center point.

After selecting the prior box, SoftMax is used to classify if it is foreground or not and bounding box regression is used to get object location.

Page updated

Google Sites

Report abuse