|
1
|
|
|
# coding=utf-8 |
|
2
|
|
|
|
|
3
|
|
|
""" |
|
4
|
|
|
Tests for deepreg/model/loss/label.py in |
|
5
|
|
|
pytest style |
|
6
|
|
|
""" |
|
7
|
|
|
from test.unit.util import is_equal_tf |
|
8
|
|
|
from typing import List, Optional, Union |
|
9
|
|
|
|
|
10
|
|
|
import numpy as np |
|
11
|
|
|
import pytest |
|
12
|
|
|
import tensorflow as tf |
|
13
|
|
|
|
|
14
|
|
|
from deepreg.loss.label import DiceLoss, DiceScore |
|
15
|
|
|
from deepreg.loss.util import MultiScaleMixin, separable_filter |
|
16
|
|
|
|
|
17
|
|
|
|
|
18
|
|
|
class TestMultiScaleMixin: |
|
19
|
|
|
def test_err(self): |
|
20
|
|
|
with pytest.raises(ValueError) as err_info: |
|
21
|
|
|
MultiScaleMixin(kernel="unknown") |
|
22
|
|
|
assert "Kernel unknown is not supported." in str(err_info.value) |
|
23
|
|
|
|
|
24
|
|
|
def test_get_config(self): |
|
25
|
|
|
loss = MultiScaleMixin() |
|
26
|
|
|
got = loss.get_config() |
|
27
|
|
|
expected = dict( |
|
28
|
|
|
scales=None, |
|
29
|
|
|
kernel="gaussian", |
|
30
|
|
|
reduction=tf.keras.losses.Reduction.AUTO, |
|
31
|
|
|
name=None, |
|
32
|
|
|
) |
|
33
|
|
|
assert got == expected |
|
34
|
|
|
|
|
35
|
|
|
@pytest.mark.parametrize("kernel", ["gaussian", "cauchy"]) |
|
36
|
|
|
@pytest.mark.parametrize("scales", [None, 0, [0], [0, 1], [1, 2]]) |
|
37
|
|
|
def test_call(self, kernel: str, scales: Optional[Union[List, float, int]]): |
|
38
|
|
|
""" |
|
39
|
|
|
Test MultiScaleMixin using DiceLoss. |
|
40
|
|
|
|
|
41
|
|
|
:param kernel: kernel name. |
|
42
|
|
|
:param scales: scaling parameters. |
|
43
|
|
|
""" |
|
44
|
|
|
shape = (2, 3, 4, 5) |
|
45
|
|
|
y_true = tf.random.uniform(shape=shape) |
|
46
|
|
|
y_pred = tf.random.uniform(shape=shape) |
|
47
|
|
|
|
|
48
|
|
|
loss = DiceLoss(kernel=kernel, scales=scales) |
|
49
|
|
|
loss.call(y_pred=y_pred, y_true=y_true) |
|
50
|
|
|
|
|
51
|
|
|
|
|
52
|
|
|
def test_negative_loss_mixin(): |
|
53
|
|
|
"""Test DiceScore and DiceLoss have reversed sign.""" |
|
54
|
|
|
shape = (2, 3, 4, 5) |
|
55
|
|
|
y_true = tf.random.uniform(shape=shape) |
|
56
|
|
|
y_pred = tf.random.uniform(shape=shape) |
|
57
|
|
|
|
|
58
|
|
|
dice_score = DiceScore().call(y_pred=y_pred, y_true=y_true) |
|
59
|
|
|
dice_loss = DiceLoss().call(y_pred=y_pred, y_true=y_true) |
|
60
|
|
|
|
|
61
|
|
|
assert is_equal_tf(dice_score, -dice_loss) |
|
62
|
|
|
|
|
63
|
|
|
|
|
64
|
|
|
def test_separable_filter(): |
|
65
|
|
|
"""Testing separable filter case where diagonal ones are propagated.""" |
|
66
|
|
|
k = tf.ones(shape=(3, 3, 3, 3, 1), dtype=tf.float32) |
|
67
|
|
|
|
|
68
|
|
|
array_eye = np.identity(3) |
|
69
|
|
|
x = np.zeros((3, 3, 3, 3, 1)) |
|
70
|
|
|
x[:, :, 0, 0, 0] = array_eye |
|
71
|
|
|
x = tf.convert_to_tensor(x, dtype=tf.float32) |
|
72
|
|
|
|
|
73
|
|
|
expected = tf.ones(shape=(3, 3, 3, 3, 1), dtype=tf.float32) |
|
74
|
|
|
got = separable_filter(x, k) |
|
75
|
|
|
|
|
76
|
|
|
assert is_equal_tf(got, expected) |
|
77
|
|
|
|
DeepRegNet /
DeepReg
