Evaluation API

Metrics

IoU (Intersection over Union)

class IoU(Metric):
    def update(self, pred, target)
    def compute() -> float

Formula:

\text{IoU} = \frac{\text{Intersection}}{\text{Union}}

F1Score (Dice Coefficient)

Harmonic mean of precision and recall.

F1 = 2 \cdot \frac{\text{precision} \cdot \text{recall}}{\text{precision} + \text{recall}}

ConfusionMatrix

Computes the \(C \times C\) confusion matrix for \(C\) classes.

Uncertainty

ExpectedCalibrationError

Measures how well predicted probabilities match accuracy.

\text{ECE} = \sum_{m=1}^M \frac{|B_m|}{N} |\text{acc}(B_m) - \text{conf}(B_m)|