def segmentation(num_classes, ignore_idx=255, attach_to=None):
confusion_matrix = metrics.ConfusionMatrix(num_classes=num_classes,
ignore_idx=ignore_idx,
attach_to=attach_to)
return metrics.MetricCompose(
metric_dict={
'acc': metrics.Accuracy(attach_to=attach_to),
'confusion_matrix': confusion_matrix,
'miou': metrics.mIoU(confusion_matrix)
})
def monocular_depth(attach_to=None):
return metrics.MetricCompose(
metric_dict={
'rmse':
metrics.RootMeanSquaredError(attach_to=attach_to),
'rmse_log':
metrics.RootMeanSquaredError(log_scale=True,
attach_to=attach_to),
'rmse_scale_inv':
metrics.ScaleInveriantMeanSquaredError(attach_to=attach_to),
'abs rel':
metrics.AbsoluteRelativeDifference(attach_to=attach_to),
'sq rel':
metrics.SquaredRelativeDifference(attach_to=attach_to),
'percents within thresholds':
metrics.Threshold(thresholds=[1.25, 1.25**2, 1.25**3],
attach_to=attach_to)
})
def loss_metric(loss_fn):
return metrics.MetricCompose(
metric_dict={'loss': metrics.AverageMetric(loss_fn)})
def regression(attach_to=None):
return metrics.MetricCompose(
metric_dict={'mse': metrics.MeanSquaredError(attach_to=attach_to)})
def classification(attach_to=None):
return metrics.MetricCompose(
metric_dict={'acc': metrics.Accuracy(attach_to=attach_to)})