def test_mean_iou():
pred_size = (10, 30, 30)
num_classes = 19
ignore_index = 255
results = np.random.randint(0, num_classes, size=pred_size)
label = np.random.randint(0, num_classes, size=pred_size)
label[:, 2, 5:10] = ignore_index
ret_metrics = mean_iou(results, label, num_classes, ignore_index)
all_acc, acc, iou = ret_metrics['aAcc'], ret_metrics['Acc'], ret_metrics[
'IoU']
all_acc_l, acc_l, iou_l = legacy_mean_iou(results, label, num_classes,
ignore_index)
assert all_acc == all_acc_l
assert np.allclose(acc, acc_l)
assert np.allclose(iou, iou_l)
results = np.random.randint(0, 5, size=pred_size)
label = np.random.randint(0, 4, size=pred_size)
ret_metrics = mean_iou(results,
label,
num_classes,
ignore_index=255,
nan_to_num=-1)
all_acc, acc, iou = ret_metrics['aAcc'], ret_metrics['Acc'], ret_metrics[
'IoU']
assert acc[-1] == -1
assert acc[-1] == -1
def get_evaluations(log_proba, seg, n_classes, ignore_idx):
res = log_proba.argmax(1)
miou, cate_acc, cate_iou = mean_iou(results=res,
gt_seg_maps=seg,
num_classes=n_classes,
ignore_index=ignore_idx,
nan_to_num=0.0)
# record
return miou, cate_acc, cate_iou
def test_mean_iou():
pred_size = (10, 30, 30)
num_classes = 19
ignore_index = 255
results = np.random.randint(0, num_classes, size=pred_size)
label = np.random.randint(0, num_classes, size=pred_size)
label[:, 2, 5:10] = ignore_index
all_acc, acc, iou = mean_iou(results, label, num_classes, ignore_index)
all_acc_l, acc_l, iou_l = legacy_mean_iou(results, label, num_classes,
ignore_index)
assert all_acc == all_acc_l
assert np.allclose(acc, acc_l)
assert np.allclose(iou, iou_l)
