CAccuracyLayer Class

This class implements a layer that calculates classification accuracy, that is, the proportion of objects classified correctly in the set.

Settings

void SetReset( bool value );

Specifies if the data should be reset after each network run. By default, the reset is turned on.

If you turn off this setting, the total accuracy since the last reset will be calculated.

This layer has no trainable parameters.

The layer has two inputs. The first input accepts a blob with the network response, of the dimensions:

BatchLength * BatchWidth * ListSize is equal to the number of objects that were classified.
Height, Width, and Depth are equal to 1.
Channels is equal to 1 for binary classification and to the number of classes if there are more than 2.

The second input should contain a blob with the correct class labels:

If first input Channels is equal to 1, the labels for the binary classification should contain 1 for one class and -1 for the other.
If Channels is greater than 1 for the multiple labels classification two forms are allowed:
- labels should contain a blob of the same dimensions with 1 for correct class and 0 for the others.
- labels should contain a blob with Channels = 1 with correct label indexes. Both types CT_Float and CT_Int are allowed.

The single output returns a blob with only one element, which contains the proportion of correctly classified objects among all objects.

If you have set SetReset() to false, the layer will accumulate the data for all network runs since the last reset.