def build_input_vector(face):
(gir_face, depth_face) = (face.gir_img, face.depth_img)
if gir_face is None or depth_face is None:
return None
if np.isnan(gir_face).any() or np.isnan(depth_face).any():
return None
try:
face = normalized(face, rotate=False)
face = hog_and_entropy(face)
except ValueError:
return None
return face.get_fd_desc()
person_photos_count = 0
for j in range(database.imgs_per_subject(i)):
print(i, j)
if len(samples) >= limit:
return samples
x = database.load_gird_face(i, j)
if x.gir_img is None or x.depth_img is None:
print(x.gir_img, x.depth_img)
continue
samples.append(x)
person_photos_count += 1
if limit_per_person is not None\
and person_photos_count >= limit_per_person:
break
return samples
# Load a random photo to rotate
helper = DBHelper(load_depth=True, load_ir=True, load_png=False)
TOTAL_SUBJECTS_COUNT = helper.all_subjects_count()
photos = []
for database in helper.get_databases():
if database.get_name() == 'photos':
photos += load_samples(database, limit=10, limit_per_person=3)
for face in photos:
face = normalized(face)
face = hog_and_entropy(face)
inp = face.get_concat()
show_image(inp)