def test_save_float_matrix(self):
ds.save_matrix(self.test_matrix,
os.path.join(BASE_PATH, "testdata", "writetest.mat"))
loaded = ds.load_matrix(
os.path.join(BASE_PATH, "testdata", "writetest.mat"))
self.assertTrue(
(self.test_matrix == loaded).all(),
"Should be:\n%s\nbut is:\n%s" % (self.test_matrix, loaded))
def _stack_bow_vectors(self, bow_files):
# log("Loading matrices")
bow_vectors = [storage.load_matrix(os.path.join(self.PATHS["BOW"], f)) for f in bow_files]
# log("Building sample matrix")
# build a n_samples x n_features matrix for sklearn by concatenating the BoW-vectors
matrix = np.vstack(bow_vectors)
# log("Vocabulary size: %d" % matrix.shape[1])
return matrix
def heatmap_data(self, img_name, importances, img_size):
"""Creates a 2-dimensional float array, representing
the heatmap of importances for the given image.
The code to calculate a keypoint's environment is based on
OpenCV's SIFT-implementation.
Pixels that are contained in the environments of two or
more distinct keypoints get assigned the maximum of
the possible importances.
Args:
img_name: Filename of the image in question.
importances: Array of feature importances.
img_size: Tuple of (width, height) for the image.
"""
keypoints_file = util.keypoints_name(img_name)
indices_file = util.cluster_name(img_name)
keypoints = ds.keypoints_from_file(os.path.join(self.datamanager.PATHS["KEYPOINTS"], keypoints_file))
indices = [index[0] for index in ds.load_matrix(os.path.join(self.datamanager.PATHS["BOW"], indices_file))]
assert len(keypoints) == len(indices), "Should be %d, but is %d" % (len(keypoints), len(indices))
heatmap = np.zeros(img_size[0:2])
rows = heatmap.shape[0]
cols = heatmap.shape[1]
for kp, index in zip(keypoints, indices):
cosine = math.cos(math.radians(kp.angle)) / HIST_WIDTH
sine = math.sin(math.radians(kp.angle)) / HIST_WIDTH
for i in range(-RADIUS, RADIUS + 1):
for j in range(-RADIUS, RADIUS + 1):
r = kp.y + i
c = kp.x + j
c_rot = j * cosine - i * sine
r_rot = j * sine + i * cosine;
rbin = r_rot + HIST_ARRAY_WIDTH/2 - 0.5
cbin = c_rot + HIST_ARRAY_WIDTH/2 - 0.5
if (-1 < rbin < HIST_ARRAY_WIDTH) and (-1 < cbin < HIST_ARRAY_WIDTH) and (0 <= r < rows) and (0 <= c < cols):
heatmap[r, c] = self.absmax(heatmap[r, c], importances[index])
return heatmap
def test_save_uint32_matrix(self):
ds.save_matrix(self.test_matrix_uint32, os.path.join(BASE_PATH, "testdata", "writetest_uint32.mat"))
loaded = ds.load_matrix(os.path.join(BASE_PATH, "testdata", "writetest_uint32.mat"))
self.assertTrue((self.test_matrix_uint32 == loaded).all(), "Should be:\n%s\nbut is:\n%s" % (self.test_matrix_uint32, loaded))
def test_load_float_matrix(self):
loaded_matrix = ds.load_matrix(os.path.join(BASE_PATH, "testdata", "test.mat"))
self.assertTrue((self.test_matrix == loaded_matrix).all(), "Should be:\n%s\nbut is:\n%s" % (self.test_matrix, loaded_matrix))
def test_load_float_matrix(self):
loaded_matrix = ds.load_matrix(
os.path.join(BASE_PATH, "testdata", "test.mat"))
self.assertTrue(
(self.test_matrix == loaded_matrix).all(),
"Should be:\n%s\nbut is:\n%s" % (self.test_matrix, loaded_matrix))