def eval_ood_measure(conf, seg_label, pred, out_labels=[10], mask=None):
ECE = _ECELoss()
pred_ID=pred[seg_label!=out_labels[0]]
seg_label_ID = seg_label[seg_label != out_labels[0]]
conf_ID=conf[seg_label != out_labels[0]]
ece_out = ECE.forward(conf_ID,pred_ID,seg_label_ID)
conf=conf.cpu().numpy()
seg_label = seg_label.cpu().numpy()
if mask is not None:
seg_label = seg_label[mask]
out_label = seg_label == out_labels[0]
for label in out_labels:
out_label = np.logical_or(out_label, seg_label == label)
in_scores = - conf[np.logical_not(out_label)]
out_scores = - conf[out_label]
print('out_label',np.sum(out_label) , np.mean(out_scores), np.mean(in_scores))
if (len(out_scores) != 0) and (len(in_scores) != 0):
auroc, aupr, fpr = anom_utils.get_and_print_results(out_scores, in_scores)
return auroc, aupr, fpr, ece_out.cpu().item()
else:
print("This image does not contain any OOD pixels or is only OOD.",out_labels )
return None
def eval_ood_measure(conf, seg_label):
out_label = seg_label == 11
out_label = np.logical_or(out_label, seg_label == 11)
in_scores = -conf[np.logical_not(out_label)]
out_scores = -conf[out_label]
if (len(out_scores) != 0) and (len(in_scores) != 0):
auroc, aupr, fpr = anom_utils.get_and_print_results(
out_scores, in_scores)
return auroc, aupr, fpr
else:
print("This image does not contain any OOD pixels or is only OOD.")
return None
def eval_ood_measure(conf, seg_label, cfg, mask=None):
out_labels = cfg.OOD.out_label
if mask is not None:
seg_label = seg_label[mask]
out_label = seg_label == out_labels[0]
for label in out_labels:
out_label = np.logical_or(out_label, seg_label == label)
in_scores = - conf[np.logical_not(out_label))]
out_scores = - conf[out_label]
if (len(out_scores) != 0) and (len(in_scores) != 0):
auroc, aupr, fpr = anom_utils.get_and_print_results(out_scores, in_scores)
return auroc, aupr, fpr
else:
print("This image does not contain any OOD pixels or is only OOD.")
return None