def test_color_jitter(self):
jitter_param = {
'brightness_factor': torch.tensor(1.),
'contrast_factor': torch.tensor(1.),
'saturation_factor': torch.tensor(1.),
'hue_factor': torch.tensor(0.),
'order': torch.tensor([2, 3, 0, 1])
}
input = torch.rand(3, 5, 5) # 3 x 5 x 5
expected = input
expected_transform = torch.eye(3).unsqueeze(0) # 3 x 3
assert_allclose(F.apply_color_jitter(input, jitter_param),
expected,
atol=1e-4,
rtol=1e-5)
assert_allclose(F.apply_color_jitter(input,
jitter_param,
return_transform=True)[0],
expected,
atol=1e-4,
rtol=1e-5)
assert_allclose(
F.apply_color_jitter(input, jitter_param,
return_transform=True)[1], expected_transform)
def test_color_jitter_batch(self):
batch_size = 2
jitter_param = {
'brightness_factor': torch.tensor([1.] * batch_size),
'contrast_factor': torch.tensor([1.] * batch_size),
'saturation_factor': torch.tensor([1.] * batch_size),
'hue_factor': torch.tensor([0.] * batch_size),
'order': torch.tensor([2, 3, 0, 1])
}
input = torch.rand(batch_size, 3, 5, 5) # 2 x 3 x 5 x 5
expected = input
expected_transform = torch.eye(3).unsqueeze(0).expand(
(batch_size, 3, 3)) # 2 x 3 x 3
assert_allclose(F.apply_color_jitter(input, jitter_param),
expected,
atol=1e-4,
rtol=1e-5)
assert_allclose(F.apply_color_jitter(input,
jitter_param,
return_transform=True)[0],
expected,
atol=1e-4,
rtol=1e-5)
assert_allclose(
F.apply_color_jitter(input, jitter_param,
return_transform=True)[1], expected_transform)
def test_random_brightness(self):
torch.manual_seed(42)
jitter_param = {
'brightness_factor': torch.tensor([1.1529, 1.1660]),
'contrast_factor': torch.tensor([1., 1.]),
'hue_factor': torch.tensor([0., 0.]),
'saturation_factor': torch.tensor([1., 1.]),
'order': torch.tensor([2, 3, 0, 1])
}
input = torch.tensor([[[[0.1, 0.2, 0.3], [0.6, 0.5, 0.4],
[0.7, 0.8, 1.]]]]) # 1 x 1 x 3 x 3
input = input.repeat(2, 3, 1, 1) # 2 x 3 x 3
expected = torch.tensor([[[[0.2529, 0.3529, 0.4529],
[0.7529, 0.6529, 0.5529],
[0.8529, 0.9529, 1.0000]],
[[0.2529, 0.3529, 0.4529],
[0.7529, 0.6529, 0.5529],
[0.8529, 0.9529, 1.0000]],
[[0.2529, 0.3529, 0.4529],
[0.7529, 0.6529, 0.5529],
[0.8529, 0.9529, 1.0000]]],
[[[0.2660, 0.3660, 0.4660],
[0.7660, 0.6660, 0.5660],
[0.8660, 0.9660, 1.0000]],
[[0.2660, 0.3660, 0.4660],
[0.7660, 0.6660, 0.5660],
[0.8660, 0.9660, 1.0000]],
[[0.2660, 0.3660, 0.4660],
[0.7660, 0.6660, 0.5660],
[0.8660, 0.9660,
1.0000]]]]) # 1 x 1 x 3 x 3
assert_allclose(F.apply_color_jitter(input, jitter_param), expected)
def test_random_contrast(self):
torch.manual_seed(42)
jitter_param = {
'brightness_factor': torch.tensor([1.0, 1.0]),
'contrast_factor': torch.tensor([0.9531, 1.1837]),
'hue_factor': torch.tensor([0.0, 0.0]),
'saturation_factor': torch.tensor([1.0, 1.0]),
'order': torch.tensor([2, 3, 0, 1]),
}
input = torch.tensor([[[[0.1, 0.2, 0.3], [0.6, 0.5, 0.4], [0.7, 0.8, 1.0]]]]) # 1 x 1 x 3 x 3
input = input.repeat(2, 3, 1, 1) # 2 x 3 x 3
expected = torch.tensor(
[
[
[[0.0953, 0.1906, 0.2859], [0.5719, 0.4766, 0.3813], [0.6672, 0.7625, 0.9531]],
[[0.0953, 0.1906, 0.2859], [0.5719, 0.4766, 0.3813], [0.6672, 0.7625, 0.9531]],
[[0.0953, 0.1906, 0.2859], [0.5719, 0.4766, 0.3813], [0.6672, 0.7625, 0.9531]],
],
[
[[0.1184, 0.2367, 0.3551], [0.7102, 0.5919, 0.4735], [0.8286, 0.9470, 1.0000]],
[[0.1184, 0.2367, 0.3551], [0.7102, 0.5919, 0.4735], [0.8286, 0.9470, 1.0000]],
[[0.1184, 0.2367, 0.3551], [0.7102, 0.5919, 0.4735], [0.8286, 0.9470, 1.0000]],
],
]
)
assert_allclose(F.apply_color_jitter(input, jitter_param), expected, atol=1e-4, rtol=1e-5)
def test_color_jitter_batch(self):
batch_size = 2
jitter_param = {
'brightness_factor': torch.tensor([1.0] * batch_size),
'contrast_factor': torch.tensor([1.0] * batch_size),
'saturation_factor': torch.tensor([1.0] * batch_size),
'hue_factor': torch.tensor([0.0] * batch_size),
'order': torch.tensor([2, 3, 0, 1]),
}
input = torch.rand(batch_size, 3, 5, 5) # 2 x 3 x 5 x 5
expected = input
assert_allclose(F.apply_color_jitter(input, jitter_param), expected, atol=1e-4, rtol=1e-5)
def test_random_saturation(self):
torch.manual_seed(42)
jitter_param = {
'brightness_factor': torch.tensor([1., 1.]),
'contrast_factor': torch.tensor([1., 1.]),
'hue_factor': torch.tensor([0., 0.]),
'saturation_factor': torch.tensor([0.9026, 1.1175]),
'order': torch.tensor([2, 3, 0, 1])
}
input = torch.tensor([[[[0.1, 0.2, 0.3],
[0.6, 0.5, 0.4],
[0.7, 0.8, 1.]],
[[1.0, 0.5, 0.6],
[0.6, 0.3, 0.2],
[0.8, 0.1, 0.2]],
[[0.6, 0.8, 0.7],
[0.9, 0.3, 0.2],
[0.8, 0.4, .5]]]]) # 1 x 1 x 3 x 3
input = input.repeat(2, 1, 1, 1) # 2 x 3 x 3
expected = torch.tensor([[[[1.8763e-01, 2.5842e-01, 3.3895e-01],
[6.2921e-01, 5.0000e-01, 4.0000e-01],
[7.0974e-01, 8.0000e-01, 1.0000e+00]],
[[1.0000e+00, 5.2921e-01, 6.0974e-01],
[6.2921e-01, 3.1947e-01, 2.1947e-01],
[8.0000e-01, 1.6816e-01, 2.7790e-01]],
[[6.3895e-01, 8.0000e-01, 7.0000e-01],
[9.0000e-01, 3.1947e-01, 2.1947e-01],
[8.0000e-01, 4.3895e-01, 5.4869e-01]]],
[[[1.1921e-07, 1.2953e-01, 2.5302e-01],
[5.6476e-01, 5.0000e-01, 4.0000e-01],
[6.8825e-01, 8.0000e-01, 1.0000e+00]],
[[1.0000e+00, 4.6476e-01, 5.8825e-01],
[5.6476e-01, 2.7651e-01, 1.7651e-01],
[8.0000e-01, 1.7781e-02, 1.0603e-01]],
[[5.5556e-01, 8.0000e-01, 7.0000e-01],
[9.0000e-01, 2.7651e-01, 1.7651e-01],
[8.0000e-01, 3.5302e-01, 4.4127e-01]]]])
assert_allclose(F.apply_color_jitter(input, jitter_param), expected, atol=1e-4, rtol=1e-5)
def test_color_jitter(self):
jitter_param = {
'brightness_factor': torch.tensor(1.0),
'contrast_factor': torch.tensor(1.0),
'saturation_factor': torch.tensor(1.0),
'hue_factor': torch.tensor(0.0),
'order': torch.tensor([2, 3, 0, 1]),
}
input = torch.rand(3, 5, 5) # 3 x 5 x 5
expected = input
assert_allclose(F.apply_color_jitter(input[None], jitter_param), expected[None], atol=1e-4, rtol=1e-5)
def test_random_hue(self):
torch.manual_seed(42)
jitter_param = {
'brightness_factor': torch.tensor([1., 1.]),
'contrast_factor': torch.tensor([1., 1.]),
'hue_factor': torch.tensor([-0.0438 / 2 / pi, 0.0404 / 2 / pi]),
'saturation_factor': torch.tensor([1., 1.]),
'order': torch.tensor([2, 3, 0, 1])
}
input = torch.tensor([[[[0.1, 0.2, 0.3],
[0.6, 0.5, 0.4],
[0.7, 0.8, 1.]],
[[1.0, 0.5, 0.6],
[0.6, 0.3, 0.2],
[0.8, 0.1, 0.2]],
[[0.6, 0.8, 0.7],
[0.9, 0.3, 0.2],
[0.8, 0.4, .5]]]]) # 1 x 1 x 3 x 3
input = input.repeat(2, 1, 1, 1) # 2 x 3 x 3
expected = torch.tensor([[[[0.1000, 0.2000, 0.3000],
[0.6000, 0.5000, 0.4000],
[0.7000, 0.8000, 1.0000]],
[[1.0000, 0.5251, 0.6167],
[0.6126, 0.3000, 0.2000],
[0.8000, 0.1000, 0.2000]],
[[0.5623, 0.8000, 0.7000],
[0.9000, 0.3084, 0.2084],
[0.7958, 0.4293, 0.5335]]],
[[[0.1000, 0.2000, 0.3000],
[0.6116, 0.5000, 0.4000],
[0.7000, 0.8000, 1.0000]],
[[1.0000, 0.4769, 0.5846],
[0.6000, 0.3077, 0.2077],
[0.7961, 0.1000, 0.2000]],
[[0.6347, 0.8000, 0.7000],
[0.9000, 0.3000, 0.2000],
[0.8000, 0.3730, 0.4692]]]])
assert_allclose(F.apply_color_jitter(input, jitter_param), expected, atol=1e-4, rtol=1e-5)
