def testCreateImage(self):
with TempFile(suffix = '.jpg') as path:
toimage(np.random.random((256, 128)) * 255).save(path)
source = InputSource(path)
img = source.CreateImage()
self.assertIsNotNone(img)
self.assertEqual(img.size, (128, 256))
self.assertEqual(img.fp.name, path)
def testExtractFeatures_multipleImages(self):
m = Model()
img1 = toimage(np.random.random((100, 100)) * 255)
img2 = toimage(np.random.random((100, 100)) * 255)
s1 = BuildLayer(m, Layer.C1, m.MakeState(img1))
s2 = BuildLayer(m, Layer.C1, m.MakeState(img2))
fs = ExtractFeatures(Layer.S1, [s1, s2])
self.assertEqual(fs.shape[0], 2)
self.assertGreater(fs.shape[1], 0)
def testExtractFeatures_multipleImages(self):
m = Model()
img1 = toimage(np.random.random((100, 100)) * 255)
img2 = toimage(np.random.random((100, 100)) * 255)
s1 = BuildLayer(m, Layer.C1, m.MakeState(img1))
s2 = BuildLayer(m, Layer.C1, m.MakeState(img2))
fs = ExtractFeatures(Layer.S1, [s1, s2])
self.assertEqual(fs.shape[0], 2)
self.assertGreater(fs.shape[1], 0)
def testComputeActivationMaps_multipleLayers(self):
m = Model()
img1 = toimage(np.random.random((100,100)) * 255)
img2 = toimage(np.random.random((100,100)) * 255)
exp = ExperimentData()
exp.corpus.paths = [img1, img2]
exp.extractor.model = Model()
ComputeActivation(exp, Layer.C1, SinglecorePool(), save_all=True)
states = exp.extractor.activation
self.assertEqual(len(states), 2)
for state in states:
self.assertEqual(len(state), 4)
for layer in (Layer.IMAGE, Layer.RETINA, Layer.S1, Layer.C1):
self.assertIn(layer, state)
def testComputeActivationMaps_multipleLayers(self):
m = Model()
img1 = toimage(np.random.random((100, 100)) * 255)
img2 = toimage(np.random.random((100, 100)) * 255)
exp = ExperimentData()
exp.corpus.paths = [img1, img2]
exp.extractor.model = Model()
ComputeActivation(exp, Layer.C1, SinglecorePool(), save_all=True)
states = exp.extractor.activation
self.assertEqual(len(states), 2)
for state in states:
self.assertEqual(len(state), 4)
for layer in (Layer.IMAGE, Layer.RETINA, Layer.S1, Layer.C1):
self.assertIn(layer, state)
def test_noResize(self): data = np.random.random((100,101)) * 255 data = data.astype(np.uint8) img = toimage(data) params = Params(image_resize_method = ResizeMethod.NONE) img2 = PrepareImage(img, params) self.assertEqual(img, img2)
def testExtractFeatures_multipleLayers(self):
m = Model()
img = toimage(np.random.random((100, 100)) * 255)
s = BuildLayer(m, Layer.C1, m.MakeState(img))
fs = ExtractFeatures((Layer.S1, Layer.C1), [s])
self.assertEqual(fs.shape[0], 1)
self.assertGreater(fs.shape[1], 0)
def testExtractFeatures_multipleLayers(self):
m = Model()
img = toimage(np.random.random((100, 100)) * 255)
s = BuildLayer(m, Layer.C1, m.MakeState(img))
fs = ExtractFeatures((Layer.S1, Layer.C1), [s])
self.assertEqual(fs.shape[0], 1)
self.assertGreater(fs.shape[1], 0)
def testBuild_C1FromImage(self):
model = Model(Params(retina_enabled=True))
img = np.random.random((100,100)).astype(ACTIVATION_DTYPE)
state = BuildLayer(model, Layer.C1, model.MakeState(toimage(img)))
for layer in (Layer.IMAGE, Layer.RETINA, Layer.S1, Layer.C1):
self.assertIn(layer, state)
self.assertIsNotNone(state[layer])
def testBuild_C1FromImage(self):
model = Model(Params(retina_enabled=True))
img = np.random.random((100, 100)).astype(ACTIVATION_DTYPE)
state = BuildLayer(model, Layer.C1, model.MakeState(toimage(img)))
for layer in (Layer.IMAGE, Layer.RETINA, Layer.S1, Layer.C1):
self.assertIn(layer, state)
self.assertIsNotNone(state[layer])
def test_noResize(self):
data = np.random.random((100, 101)) * 255
data = data.astype(np.uint8)
img = toimage(data)
params = Params(image_resize_method=ResizeMethod.NONE)
img2 = PrepareImage(img, params)
self.assertEqual(img, img2)
def _test(self, num_kernels, num_images, normalize):
p = Params(num_scales=3)
model = ReshapedModel(p)
kernel_sizes = (5, 7, 11)
states = [
model.MakeState(toimage(np.random.random((100, 100))))
for _ in range(num_images)
]
kernels, locs = SamplePatchesFromImages(model,
ReshapedLayer.RESHAPED_IMAGE,
kernel_sizes,
num_kernels,
states,
SinglecorePool(),
normalize=normalize)
self.assertEqual(len(kernels), len(kernel_sizes))
self.assertEqual(len(locs), len(kernel_sizes))
for idx, ksize in enumerate(kernel_sizes):
self.assertEqual(len(kernels[idx]), num_kernels)
self.assertEqual(len(locs[idx]), num_kernels)
for k, l in zip(kernels[idx], locs[idx]):
self.assertEqual(k.ndim, 2)
self.assertEqual(len(l), 4)
if normalize:
self.assertAlmostEqual(
np.linalg.norm(k), 1, places=5
) # don't be too strict with floating-point comparison
def testComputeActivationMaps_noSaveAll(self): m = Model() img = toimage(np.random.random((100,100)) * 255) exp = ExperimentData() exp.corpus.paths = [img] exp.extractor.model = Model() ComputeActivation(exp, Layer.C1, SinglecorePool(), save_all=False) states = exp.extractor.activation self.assertEqual(len(states), 1) self.assertEqual(len(states[0]), 1) self.assertIn(Layer.C1, states[0])
def testComputeActivationMaps_noSaveAll(self):
m = Model()
img = toimage(np.random.random((100, 100)) * 255)
exp = ExperimentData()
exp.corpus.paths = [img]
exp.extractor.model = Model()
ComputeActivation(exp, Layer.C1, SinglecorePool(), save_all=False)
states = exp.extractor.activation
self.assertEqual(len(states), 1)
self.assertEqual(len(states[0]), 1)
self.assertIn(Layer.C1, states[0])
def ToImage(data):
"""Create an image from a 2D array of model activity.
:param data: Single scale of one layer of model activity, with elements in the
range [0,1].
:type data: 2D ndarray of floats
:rtype: Image
:returns: Greyscale image of layer activity.
"""
if not (isinstance(data, np.ndarray) and data.ndim == 2 and \
data.dtype == ACTIVATION_DTYPE):
raise ValueError("Invalid image layer data")
data = data.copy()
data *= 255
img = toimage(data.astype(np.uint8))
return img
def ToImage(data):
"""Create an image from a 2D array of model activity.
:param data: Single scale of one layer of model activity, with elements in the
range [0,1].
:type data: 2D ndarray of floats
:rtype: Image
:returns: Greyscale image of layer activity.
"""
if not (isinstance(data, np.ndarray) and data.ndim == 2 and \
data.dtype == ACTIVATION_DTYPE):
raise ValueError("Invalid image layer data")
data = data.copy()
data *= 255
img = toimage(data.astype(np.uint8))
return img
def _test(self, num_kernels, num_images, normalize):
p = Params(num_scales = 3)
model = ReshapedModel(p)
kernel_sizes = (5, 7, 11)
states = [ model.MakeState(toimage(np.random.random((100,100))))
for _ in range(num_images) ]
kernels, locs = SamplePatchesFromImages(model, ReshapedLayer.RESHAPED_IMAGE,
kernel_sizes, num_kernels, states, SinglecorePool(),
normalize = normalize)
self.assertEqual(len(kernels), len(kernel_sizes))
self.assertEqual(len(locs), len(kernel_sizes))
for idx, ksize in enumerate(kernel_sizes):
self.assertEqual(len(kernels[idx]), num_kernels)
self.assertEqual(len(locs[idx]), num_kernels)
for k, l in zip(kernels[idx], locs[idx]):
self.assertEqual(k.ndim, 2)
self.assertEqual(len(l), 4)
if normalize:
self.assertAlmostEqual(np.linalg.norm(k), 1,
places = 5) # don't be too strict with floating-point comparison
