def test_one_cls_dice_one_cls(self):
batch = 5
class_num = 2 # class type: 0,1,2,3
x = 10
y = 10
z = 10
output = np.random.rand(batch, class_num, x, y, z)
target = np.random.randint(0, 2, (batch, x, y, z), 'int')
v_output = Variable(torch.Tensor(output))
v_target = Variable(torch.Tensor(target))
dice1 = one_cls_dice(v_output, v_target, 1)
dice0 = one_cls_dice(v_output, v_target, 0)
assert dice1 < 1
assert dice0 < 1
def dice_cat2(output, target):
"""
Calculate dice for all channels.
:param output:Output dimension: Batch x Channel x X x Y (x Z) float
:param target:Target dimension: Batch x X x Y (x Z) int:[0, Channel]
:return:
"""
return one_cls_dice(output, target, label_idx=2)
def test_one_cls_dice_identical(self):
batch = 5
class_num = 2 # class type: 0,1,2,3
x = 10
y = 10
z = 10
output = np.random.rand(batch, class_num, x, y, z)
target = np.argmax(output, 1)
v_output = Variable(torch.Tensor(output))
v_target = Variable(torch.Tensor(target))
dice1 = one_cls_dice(v_output, v_target, 1)
dice2 = one_cls_dice(v_output, v_target, 2)
dice3 = one_cls_dice(v_output, v_target, 3)
dice0 = one_cls_dice(v_output, v_target, 0)
assert math.isclose(dice1, 1, rel_tol=0.001)
assert math.isclose(dice2, 0, rel_tol=0.001)
assert math.isclose(dice3, 0, rel_tol=0.001)
assert math.isclose(dice0, 1, rel_tol=0.001)
class_num = 4
output = np.random.rand(batch, class_num, x, y, z)
target = np.argmax(output, 1)
v_output = Variable(torch.Tensor(output))
v_target = Variable(torch.Tensor(target))
dice1 = one_cls_dice(v_output, v_target, 1)
dice2 = one_cls_dice(v_output, v_target, 2)
dice3 = one_cls_dice(v_output, v_target, 3)
dice0 = one_cls_dice(v_output, v_target, 0)
assert math.isclose(dice1, 1, rel_tol=0.001)
assert math.isclose(dice2, 1, rel_tol=0.001)
assert math.isclose(dice3, 1, rel_tol=0.001)
assert math.isclose(dice0, 1, rel_tol=0.001)