Skip to content

SaigeClassRoc

  • #include <SaigeVisionAPI.h>

ROC curve and AUROC of a single class, computed during validation at training time.

The lengths of fpr, tpr, and threshold are identical (numPoints), but may differ across classes. For SEG models these are pixel-level metrics.

Public Attributes

Type Name
int classIndex
Class index this ROC belongs to, matching SaigeModelInfo::classNames. Note that classes without a metric (e.g. the OK class of SEG) are omitted, so indices may be non-contiguous.
float auroc
Area under the ROC curve.
int numPoints
Number of points in the fpr, tpr, and threshold arrays.
const float * fpr
False positive rates, length numPoints.
const float * tpr
True positive rates, length numPoints.
const float * threshold
Thresholds, length numPoints.

Public Attributes Documentation

classIndex

Class index this ROC belongs to, matching SaigeModelInfo::classNames. Note that classes without a metric (e.g. the OK class of SEG) are omitted, so indices may be non-contiguous.

int classIndex;

auroc

Area under the ROC curve.

float auroc;

numPoints

Number of points in the fpr, tpr, and threshold arrays.

int numPoints;

fpr

False positive rates, length numPoints.

const float * fpr;

tpr

True positive rates, length numPoints.

const float * tpr;

threshold

Thresholds, length numPoints.

const float * threshold;