December 2018
Intermediate to advanced
764 pages
18h 18m
English
Metrics are essential for understanding in the context of a deep learning task. The metrics of object detection and localization are peculiar because of human annotation. The human may have annotated a box that is called ground-truth. The ground-truth need not be the absolute truth. Moreover, the boxes can be a few pixels different from human to human. Hence it becomes harder for the algorithm to detect the exact bounding box drawn by humans. Intersection over Union (IoU) is used to evaluate the localization task. Mean Precision Average (mAP) is used to evaluate the detection task. We will see the descriptions of the metrics in the next sections.
Read now
Unlock full access