def test_opencv_false_batch(self, device):
batch_size = 4
grayscale_params = {'batch_prob': torch.tensor([False] * batch_size)}
data = torch.tensor(
[[[0.3944633, 0.8597369, 0.1670904, 0.2825457, 0.0953912],
[0.1251704, 0.8020709, 0.8933256, 0.9170977, 0.1497008],
[0.2711633, 0.1111478, 0.0783281, 0.2771807, 0.5487481],
[0.0086008, 0.8288748, 0.9647092, 0.8922020, 0.7614344],
[0.2898048, 0.1282895, 0.7621747, 0.5657831, 0.9918593]],
[[0.5414237, 0.9962701, 0.8947155, 0.5900949, 0.9483274],
[0.0468036, 0.3933847, 0.8046577, 0.3640994, 0.0632100],
[0.6171775, 0.8624780, 0.4126036, 0.7600935, 0.7279997],
[0.4237089, 0.5365476, 0.5591233, 0.1523191, 0.1382165],
[0.8932794, 0.8517839, 0.7152701, 0.8983801, 0.5905426]],
[[0.2869580, 0.4700376, 0.2743714, 0.8135023, 0.2229074],
[0.9306560, 0.3734594, 0.4566821, 0.7599275, 0.7557513],
[0.7415742, 0.6115875, 0.3317572, 0.0379378, 0.1315770],
[0.8692724, 0.0809556, 0.7767404, 0.8742208, 0.1522012],
[0.7708948, 0.4509611, 0.0481175, 0.2358997, 0.6900532]]])
data = data.to(device)
data = data.unsqueeze(0).repeat(batch_size, 1, 1, 1)
expected = data
assert_allclose(F._apply_grayscale(data, grayscale_params), expected)
def test_random_grayscale_return_transform(self):
grayscale_params_0 = {'batch_prob': torch.tensor([True])}
grayscale_params_1 = {'batch_prob': torch.tensor([False])}
input = torch.rand(3, 5, 5) # 3 x 5 x 5
expected_transform = torch.eye(3).unsqueeze(0) # 3 x 3
assert_allclose(
F._apply_grayscale(input,
grayscale_params_0,
return_transform=True)[1], expected_transform)
assert_allclose(
F._apply_grayscale(input,
grayscale_params_1,
return_transform=True)[1], expected_transform)
def test_opencv_true_batch(self, device):
batch_size = 4
grayscale_params = {'batch_prob': torch.tensor([True] * batch_size)}
data = torch.tensor(
[[[0.3944633, 0.8597369, 0.1670904, 0.2825457, 0.0953912],
[0.1251704, 0.8020709, 0.8933256, 0.9170977, 0.1497008],
[0.2711633, 0.1111478, 0.0783281, 0.2771807, 0.5487481],
[0.0086008, 0.8288748, 0.9647092, 0.8922020, 0.7614344],
[0.2898048, 0.1282895, 0.7621747, 0.5657831, 0.9918593]],
[[0.5414237, 0.9962701, 0.8947155, 0.5900949, 0.9483274],
[0.0468036, 0.3933847, 0.8046577, 0.3640994, 0.0632100],
[0.6171775, 0.8624780, 0.4126036, 0.7600935, 0.7279997],
[0.4237089, 0.5365476, 0.5591233, 0.1523191, 0.1382165],
[0.8932794, 0.8517839, 0.7152701, 0.8983801, 0.5905426]],
[[0.2869580, 0.4700376, 0.2743714, 0.8135023, 0.2229074],
[0.9306560, 0.3734594, 0.4566821, 0.7599275, 0.7557513],
[0.7415742, 0.6115875, 0.3317572, 0.0379378, 0.1315770],
[0.8692724, 0.0809556, 0.7767404, 0.8742208, 0.1522012],
[0.7708948, 0.4509611, 0.0481175, 0.2358997, 0.6900532]]])
data = data.to(device)
data = data.unsqueeze(0).repeat(batch_size, 1, 1, 1)
# Output data generated with OpenCV 4.1.1: cv2.cvtColor(img_np, cv2.COLOR_RGB2GRAY)
expected = torch.tensor(
[[[0.4684734, 0.8954562, 0.6064363, 0.5236061, 0.6106016],
[0.1709944, 0.5133104, 0.7915002, 0.5745703, 0.1680204],
[0.5279005, 0.6092287, 0.3034387, 0.5333768, 0.6064113],
[0.3503858, 0.5720159, 0.7052018, 0.4558409, 0.3261529],
[0.6988886, 0.5897652, 0.6532392, 0.7234108, 0.7218805]],
[[0.4684734, 0.8954562, 0.6064363, 0.5236061, 0.6106016],
[0.1709944, 0.5133104, 0.7915002, 0.5745703, 0.1680204],
[0.5279005, 0.6092287, 0.3034387, 0.5333768, 0.6064113],
[0.3503858, 0.5720159, 0.7052018, 0.4558409, 0.3261529],
[0.6988886, 0.5897652, 0.6532392, 0.7234108, 0.7218805]],
[[0.4684734, 0.8954562, 0.6064363, 0.5236061, 0.6106016],
[0.1709944, 0.5133104, 0.7915002, 0.5745703, 0.1680204],
[0.5279005, 0.6092287, 0.3034387, 0.5333768, 0.6064113],
[0.3503858, 0.5720159, 0.7052018, 0.4558409, 0.3261529],
[0.6988886, 0.5897652, 0.6532392, 0.7234108, 0.7218805]]])
expected = expected.to(device)
expected = expected.unsqueeze(0).repeat(batch_size, 1, 1, 1)
assert_allclose(F._apply_grayscale(data, grayscale_params), expected)