Hands-On Convolutional Neural Networks with TensorFlow

Loss Part 3

The classification loss is the last part of loss function.

This loss is the sum of squared error loss for classification. Again the term  is 1 when there is a object on a cell, and 0 otherwise. The idea is that we don't take into account the classification error when there is on object.

The , terms serve to mask the loss on the case that we have an object on the ground-truth and have an object on the model output for a particular cell. ...