Cross entropy vs. number of iterations in the evaluation layer. The figure shows that the cross entropy loss (measures the performance of Fast RCNN Classification, that is, normal object classification) is reaching zero in the evaluation layer, thus showing that the model is achieving good classification performance.