def test_should_apply_grey_patch_on_image():
input_image = np.zeros((10, 20, 3))
output = apply_grey_patch(input_image, 0, 0, 10)
expected_output = np.concatenate(
[127.5 * np.ones(
(10, 10, 3)), np.zeros((10, 10, 3))], axis=1)
np.testing.assert_almost_equal(output, expected_output)
def get_sensitivity_map(self, model, image, class_index, patch_size):
"""
Compute sensitivity map on a given image for a specific class index.
Args:
model (tf.keras.Model): tf.keras model to inspect
image:
class_index (int): Index of targeted class
patch_size (int): Size of patch to apply on the image
Returns:
np.ndarray: Sensitivity map with shape (H, W, 3)
"""
sensitivity_map = np.zeros((
math.ceil(image.shape[0] / patch_size),
math.ceil(image.shape[1] / patch_size),
))
coordinates = [
(index_y, index_x) for index_x in range(sensitivity_map.shape[1] # pylint: disable=unsubscriptable-object
)
for index_y in range(sensitivity_map.shape[0] # pylint: disable=unsubscriptable-object
)
]
target_class_predictions = []
base_confidence = model.predict(np.array([image]),
batch_size=1)[0][class_index]
print(f"Base_confidence: {base_confidence}")
with tqdm(total=math.ceil(image.shape[0] * image.shape[1] /
patch_size**2),
position=0,
leave=True) as pbar:
for index_y, top_left_y in enumerate(
range(0, image.shape[1], patch_size)):
for index_x, top_left_x in enumerate(
range(0, image.shape[0], patch_size)):
patch = np.array([
apply_grey_patch(image, top_left_x, top_left_y,
patch_size)
])
prediction = model.predict(patch,
batch_size=1)[0][class_index]
target_class_predictions.append(prediction)
triggered = pbar.update(1)
print(f"Confidence scores: {target_class_predictions}")
for (index_y, index_x), confidence in zip(coordinates,
target_class_predictions):
sensitivity_map[index_y, index_x] = base_confidence - confidence
return cv2.resize(sensitivity_map, image.shape[0:2])
def get_sensitivity_map(self, model, image, class_index, patch_size):
"""
Compute sensitivity map on a given image for a specific class index.
Args:
model (tf.keras.Model): tf.keras model to inspect
image:
class_index (int): Index of targeted class
patch_size (int): Size of patch to apply on the image
Returns:
np.ndarray: Sensitivity map with shape (H, W, 3)
"""
sensitivity_map = np.zeros(
(
math.ceil(image.shape[0] / patch_size),
math.ceil(image.shape[1] / patch_size),
)
)
patches = [
apply_grey_patch(image, top_left_x, top_left_y, patch_size)
for index_x, top_left_x in enumerate(range(0, image.shape[0], patch_size))
for index_y, top_left_y in enumerate(range(0, image.shape[1], patch_size))
]
coordinates = [
(index_y, index_x)
for index_x in range(
sensitivity_map.shape[1] # pylint: disable=unsubscriptable-object
)
for index_y in range(
sensitivity_map.shape[0] # pylint: disable=unsubscriptable-object
)
]
predictions = model.predict(np.array(patches), batch_size=self.batch_size)
target_class_predictions = [
prediction[class_index] for prediction in predictions
]
for (index_y, index_x), confidence in zip(
coordinates, target_class_predictions
):
sensitivity_map[index_y, index_x] = 1 - confidence
return cv2.resize(sensitivity_map, image.shape[0:2])