def test_lena():
from skimage import data
from skimage import color
from skimage.transform import resize
in_shape = (200, 200)
n_imgs = 1
lena = resize(color.rgb2gray(data.lena()), in_shape).astype(np.float32)
lena -= lena.min()
lena /= lena.max()
imgs = lena.reshape((n_imgs, ) + in_shape)
model = cnnr.models.fg11_ht_l3_1_description
extractor = cnnr.BatchExtractor(in_shape, model)
feat_set = extractor.extract(imgs)
assert feat_set.shape == (n_imgs, 10, 10, 256)
feat_set.shape = n_imgs, -1
test_chunk_computed = feat_set[0, 12798:12802]
test_chunk_expected = np.array(
[0.03845372, 0.02469639, 0.01009409, 0.02500059], dtype=np.float32)
assert_allclose(test_chunk_computed,
test_chunk_expected,
rtol=RTOL,
atol=ATOL)
def test_random_images():
in_shape = (200, 200)
n_imgs = 12
model = cnnr.models.fg11_ht_l3_1_description
extractor = cnnr.BatchExtractor(in_shape, model)
rng = np.random.RandomState(42)
imgs = rng.uniform(low=0.0, high=1.0,
size=(n_imgs, ) + in_shape).astype(np.float32)
feat_set = extractor.extract(imgs)
assert feat_set.shape == (n_imgs, 10, 10, 256)
feat_set.shape = n_imgs, -1
test_chunk_computed = feat_set[3:9, 12798:12802]
test_chunk_expected = np.array(
[[0.0211054, 0.03008464, 0.01754513, 0.02833426],
[0.02208067, 0.00935635, 0.01337241, 0.01570413],
[0.009751, 0.01863629, 0.01336291, 0.01315528],
[0.01786099, 0.01888827, 0.01754452, 0.00985618],
[0.01402772, 0.01768549, 0.01073015, 0.01337401],
[0.02214634, 0.0261642, 0.00918513, 0.02125477]],
dtype=np.float32)
assert_allclose(test_chunk_computed,
test_chunk_expected,
rtol=RTOL,
atol=ATOL)
def test_no_model():
in_shape = (200, 200)
model = []
extractor = cnnr.BatchExtractor(in_shape, model)
assert extractor.n_layers == 0
assert extractor.shape_stride_by_layer == []
assert extractor.filterbanks == []
def test_ht_l3_1():
in_shape = (200, 200)
model = cnnr.models.fg11_ht_l3_1_description
extractor = cnnr.BatchExtractor(in_shape, model)
assert extractor.n_layers == 4
assert extractor.shape_stride_by_layer == [(192, 192, 9, 9, 1, 1),
(88, 88, 17, 17, 64, 2),
(34, 34, 21, 21, 128, 2),
(10, 10, 15, 15, 256, 2)]
def test_zero_images():
in_shape = (200, 200)
n_imgs = 12
model = cnnr.models.fg11_ht_l3_1_description
extractor = cnnr.BatchExtractor(in_shape, model)
imgs = np.zeros((n_imgs, ) + in_shape).astype(np.float32)
feat_set = extractor.extract(imgs)
assert feat_set.shape == (n_imgs, 10, 10, 256)
assert feat_set.sum() == 0.
