Mean IoU
Measure the intersection over union of predicted and reference masks.
Testing methods
72NLL gap
Compare negative log-likelihood across groups.
Occlusion
Hide image regions in turn and inspect the change in the model's prediction.
Partial dependence
Inspect average model predictions as a tabular feature changes.
Permutation importance
Shuffle tabular features and measure the change in model performance.
Perspective
Change image perspective, with the corresponding annotation transformation.
PGD
Test an image model with iterative projected gradient perturbations.
Pixel precision and recall
Measure pixel-level precision and recall for segmentation.
Pointing game
Check whether the strongest attribution falls inside a reference region.
R²
Compare regression residuals with variation in the reference values.
Rain
Apply simulated rain to images and compare results with the baseline.
Relative scale
Scale each tabular feature value by a configured relative amount.
RMSE
Measure the square root of the mean squared prediction error.
Saliency
Use input gradients to inspect pixel sensitivity for an image prediction.
Scale up and down
Scale images up or down, with the corresponding annotation transformation.
Snow
Apply simulated snow to images and compare results with the baseline.
Split coverage
Check whether each populated dataset split contains every declared class.
Statistical parity difference
Measure the difference in positive prediction rates across groups.
Subtractive
Subtract a configured amount from each tabular feature value.
Target completeness
Find images with no annotations or missing label files.
Target completeness on import
Validate that each imported row has a target value for scoring.
Task validation on import
Validate imported columns against the selected tabular task.
Worst-group
Identify the lowest-performing group for the selected task metric.
Zero out
Set every tabular feature value to zero and compare the model's results.
49–72 of 72 results