def test_C_adjust_mask(self):
'''
Initial Mask on line 2 and column 4 of the Ardis(64,64)
01
01
Adjust Mask with a Subwindow located in x=13(column) and y=16(line) in a double ardis (128,128)
0011
0011
0011
0011
The result should be this 4x4 mask located at line 2*2+16=20 and column 2*4+13=21
'''
fmask_gen = MaskTwoHorizontalFactory((2, 2)).next().next()[0]
fmask_gen = FeatureMask(fmask_gen.mask, (2, 4))
sw = windowgen.Subwindow(13, 16, ce=2.0)
fmask_gen.adjust_mask(sw)
mask_to_match = np.array([[0, 0, 1, 1], [0, 0, 1, 1], [0, 0, 1, 1],
[0, 0, 1, 1]])
size_to_match = (4, 4)
location_to_match = (20, 21)
fmask_match = FeatureMask(Mask(size_to_match, None), location_to_match,
mask_to_match)
assert fmask_gen == fmask_match
def test_B_ii_consistency(self):
k1 = "%s/%s" % (config.TEST_IMG_PATH, "kramer1.pgm")
k2 = "%s/%s" % (config.TEST_IMG_PATH, "kramer2.pgm")
ii_k1 = misc.IntegralImage(k1)
ii_k2 = misc.IntegralImage(k2)
fmask_k1 = MaskTwoHorizontalFactory((16, 16)).next().next()[0]
fmask_k1 = FeatureMask(fmask_k1.mask, (0, 0))
fmask_k2 = MaskTwoHorizontalFactory((16, 16)).next().next()[0]
fmask_k2 = FeatureMask(fmask_k2.mask, (0, 0))
sw_k1 = windowgen.Subwindow(80, 94)
sw_k2 = windowgen.Subwindow(0, 0)
fmask_k1.adjust_mask(sw_k1)
fmask_k2.adjust_mask(sw_k2)
assert ii_k1.filter(fmask_k1) == ii_k2.filter(fmask_k2)
def run(self):
Experiment.run(self)
# face_filenames = random.sample(os.listdir(config.FACES_PATH),self.n_img)
face_filenames = os.listdir(config.FACES_PATH)[:self.n_img]
sw = windowgen.Subwindow(0, 0, (64, 64))
true = 0
for i in face_filenames:
img_path = "%s/%s" % (config.FACES_PATH, i)
true = true + 1 if self.det.is_face(img_path, sw) else true
print "%d/%d" % (true, self.n_img)
true = 0
for scene_window in windowgen.get_next_random_image_window(
config.SCENES_PATH_VALIDATION, self.n_img):
img_path = scene_window.window_img_path
true = true + 1 if self.det.is_face(img_path, sw) else true
print "%d/%d" % (true, self.n_img)
self.stop()
def init():
folder_faces = "%s/lfwcrop_grey/faces" % (config.DATASET_PATH, )
folder_scenes = "%s/training non-face images" % (config.DATASET_PATH, )
n = 3500
T = 100
n_faces = n
n_non_faces = n
face_filenames = os.listdir(folder_faces)[:n]
faces = []
non_faces = []
'''
Create a FeatureMaster instance for each image face exemplar and puts the correspondent generator
in the appropriated list.
'''
FeatureMaster.init_training_set()
fm = FeatureMaster((64, 64), 4, 4, 1.25, 8, 8)
for face_file in face_filenames:
img_path = "%s/%s" % (folder_faces, face_file)
faces.append(img_path)
fm.add_image(img_path, FeatureMaster.FACE)
'''It does the same for the scene images'''
for scene_window in windowgen.get_next_random_image_window(
folder_scenes, n):
img_path = scene_window.window_img_path
non_faces.append(img_path)
fm.add_image(img_path, FeatureMaster.NON_FACE)
print "END STAGE 1"
fc = FeatureChooser((64, 64), fm, n_faces, n_non_faces)
classifier = fc.find_hypothesis(T)
with open('%s/%s_%d_%d.pk' % (config.CLASSIFIERS_PATH, 'classifier', n, T),
'wb') as output:
pickle.dump(classifier, output, pickle.HIGHEST_PROTOCOL)
print "END STAGE 2"
print classifier.hypothesis
#Creating the Detector
ng = 1
ref_ardis = (64, 64)
ref_mask = (8, 8)
shift_step = 4
fn = windowgen.SubwindowGenerator.FIXED_FACTOR
det = Detector(classifier, ng, ref_ardis, ref_mask, shift_step=shift_step)
true = 0
for i in range(0, n):
sw = windowgen.Subwindow(0, 0, (64, 64))
true = true + 1 if det.is_face(faces[i], sw) else true
print "%d/%d" % (true, n)
true = 0
for i in range(0, n):
sw = windowgen.Subwindow(0, 0, (64, 64))
true = true + 1 if det.is_face(non_faces[i], sw) else true
print "%d/%d" % (true, n)