Object Detection

• The confidence loss is a loss function denoted . It measures the loss on the confidence scores.

  Let be an indicator for matching the -th anchor box to the -th ground truth box

  be the confidence function of the -th default box for class

  denotes the set of positive anchor boxes (i.e., those that predicted an object class)

  denotes the set of negative anchor boxes (i.e., those that predicted the background)

  We have

  Where