def extract_img_features(self, img):
img_for_hog = pp.convert_img_for_hog(img)
hog = pp.extract_hog_features(img_for_hog)
if config.USE_SPATIAL or config.USE_COLOR_HIST:
other = pp.extract_other_features(img)
return np.concatenate((hog, other))
return hog
def test_hog_multi_channel(self):
"""
Test getting a hog layer from a single channel image.
:param self:
"""
dummy = np.zeros((64, 64, 2))
hog = pp.extract_hog_features(dummy)
self.assertGreater(len(hog), 3000)
def test_hog_single_channel(self):
"""
Test getting a hog layer from a single channel image.
:param self:
"""
dummy = np.zeros((64, 64, 1))
hog = pp.extract_hog_features(dummy)
# This number will depend on the number of blocks per cell.
self.assertGreater(len(hog), 1000)
def process_image(self, img):
'''Returns the set of feature vectors to classify from this subimage'''
img_for_hog = pp.convert_img_for_hog(img)
hogs_list = pp.extract_hog_features(img_for_hog, feature_vector=False)
hogs = np.array(hogs_list)
shape = np.shape(hogs)
window_width = shape[1]
img_width_in_blocks = shape[2]
start_col = 0
while (img_width_in_blocks - start_col >= window_width):
hogs_subsample = hogs[:, :,
start_col:start_col + window_width, :, :, ]
self.feature_windows.append((start_col, hogs_subsample.ravel()))
start_col += config.HOG_BLOCK_STEPS
def test_sliding_windows(self):
"""
Test extracting scaled regions from the image.
:param self:
"""
dummy = np.zeros((64, 64, 3), np.uint8)
img_for_hog = pp.convert_img_for_hog(dummy)
dummy_hog = pp.extract_hog_features(img_for_hog)
dummy_shape = np.shape(dummy_hog)
img = np.zeros((64, 1280, 3), np.uint8)
# just test with one channel
sliding_window = SlidingWindowHog()
sliding_window.process_image(img)
for start_col, sub_image in sliding_window.feature_windows:
self.assertEqual(np.shape(sub_image), dummy_shape)