An example of an interpretability mechanism giving information that is not fully significant is CAM or Grad-CAM [ 8 ], which can highlight an image region crucial for the classification result; however, this mechanism can not explain what, in that region, is truly important or whether it fits with the shared medical knowledge [ 9 ].
← all excerpts
Metric Learning in Histopathological Image Classification: Opening the Black Box.
1
—
—