def main():
path, img = None, None
mess = '- To index your images: ' \
'python app.py -path <path-to-your-images-directory>' \
'- To search on a certain image: ' \
'python app.py -image <path-to-your-image>'
try:
opts, _ = getopt.getopt(sys.argv[1:], 'p:i:h',
['path=', 'image=', 'help'])
except getopt.GetoptError:
print mess
sys.exit(2)
for opt, arg in opts:
if opt in ('-p', '--path'):
if os.path.isdir(arg):
path = arg
else:
print 'Path must link to a directory'
sys.exit(1)
if opt in ('-i', '--image'):
if os.path.isfile(arg):
img = arg
else:
print 'Image path must link to a image'
sys.exit(1)
if opt in ('-h', '--help'):
print mess
sys.exit(2)
if path:
print '--SOURCE PATH:' + path
prepare_db()
# Create a temporary directory to store cropped face images
faces_path = '../images/faces'
generate_faces(path, faces_path)
# Generate vectors from images and store them in database
insert_deep_features(faces_path, 'images')
if img:
print '--IMAGE PATH:' + img
# Detect the face from the image
face_img = detect_face(img)
# Get the vector from that face
vector = get_single_predictions(face_img)
# Show results
person, time = search(vector)
print '--PERSON: ' + str(person)
print '--SEARCH TIME: ' + str(time)
def frame_face_blur(image, kernel_size=20):
for face in detect_face(image):
image[face.top() - 10:face.bottom() + 10,
face.left() - 10:face.right() + 10] = pixelate(
image[face.top() - 10:face.bottom() + 10,
face.left() - 10:face.right() + 10],
kernel_size=kernel_size)
return image
def img_url_to_array_with_face_detect(url):
junk, face = face_detect.detect_face(url)
"""takes a filename and turns it into a numpy array of RGB pixels"""
img = open_image_from_url(url)
if face is not None:
dress = face_detect.dress_box(face)
img = img.crop(dress)
bbox = img.getbbox()
if bbox[2] != STANDARD_SIZE[0] or bbox[3] != STANDARD_SIZE[1]:
img = img.resize(STANDARD_SIZE)
img = list(img.getdata())
img = map(list, img)
img = np.array(img)
s = img.shape[0] * img.shape[1]
img_wide = img.reshape(1, s)
return img_wide[0]
def img_url_to_array_with_face_detect(url):
junk, face = face_detect.detect_face(url)
"""takes a filename and turns it into a numpy array of RGB pixels"""
img = open_image_from_url(url)
if face is not None:
dress = face_detect.dress_box(face)
img = img.crop(dress)
bbox = img.getbbox()
if bbox[2] != STANDARD_SIZE[0] or bbox[3] != STANDARD_SIZE[1]:
img = img.resize(STANDARD_SIZE)
img = list(img.getdata())
img = map(list, img)
img = np.array(img)
s = img.shape[0] * img.shape[1]
img_wide = img.reshape(1, s)
return img_wide[0]
def run_ref_image(image_path,
uv_kpt_ind,
face_ind,
triangles,
pnet,
rnet,
onet,
x,
y,
Tsess,
minsize=30,
threshold=[0.6, 0.7, 0.7],
factor=0.709,
best_score=0.7,
uv_h=256,
uv_w=256,
image_h=256,
image_w=256):
input_image = cv2.imread(image_path, 1)
boxes, pnts = face_detect.detect_face(input_image, minsize, pnet, rnet,
onet, threshold, factor)
faces = process_bbox(boxes, input_image.shape)
for idx, (x0, y1, x1, y0, conf_score) in enumerate(faces):
if conf_score > best_score:
det_face = input_image[int(x0):int(x1), int(y0):int(y1), :]
det_face = cv2.resize(det_face, (256, 256)) / 255.
ref_pos = Tsess.run(y,
feed_dict={x: det_face[np.newaxis, :, :, :]})
ref_pos = np.squeeze(ref_pos)
max_pos = image_h
ref_pos = ref_pos * max_pos
ref_texture = cv2.remap(det_face,
ref_pos[:, :, :2].astype(np.float32),
None,
interpolation=cv2.INTER_NEAREST,
borderMode=cv2.BORDER_CONSTANT,
borderValue=(0))
break
return ref_texture
def find_character(query):
"""Download full size images from Google image search.
Don't print or republish images without permission.
I used this to train a learning algorithm.
"""
query = query.replace('/', ' ')
path = 'tmp/characters'
BASE_PATH = path
keywords_i = 0
if os.path.exists(BASE_PATH + '/' + query + '.jpg'):
print "Reusing cached image..."
img = cv2.imread(BASE_PATH + '/' + query + '.jpg')
results = False
tries = 0
while not results and tries < 20:
print "Trying to find face again"
results = fd.detect_face(img)
if float(img.shape[1]) / float(img.shape[0]) > 1.5:
print "Need to crop %s" % query
fx, fy, fw, fh = results['face']
img = img[max(0, fy - 10):min(fy + fh + 10, img.shape[0]),
max(0, fx - 10):min(fx + fw + 10, img.shape[1])]
results = fd.detect_face(img)
if not results:
continue
tries += 1
return (results, img)
if not os.path.exists(BASE_PATH):
os.makedirs(BASE_PATH)
start = 0 # Google's start query string parameter for pagination.
while True:
if start > 3:
keywords_i += 1
start = 0
keyword = ""
character_name = ""
if keywords_i < len(character_lookup_keywords):
keyword = character_lookup_keywords[keywords_i]
character_name = query
elif (keywords_i < 2 *
len(character_lookup_keywords)) and query.lower().endswith('s'):
keyword = character_lookup_keywords[keywords_i -
len(character_lookup_keywords)]
character_name = query[:-1]
else:
character_name = 'person'
print "Searching for " + character_name + " " + keyword + " " + str(
start)
BASE_URL = 'https://ajax.googleapis.com/ajax/services/search/images?'\
'v=1.0&q=' + character_name + '+' + keyword + '&start=%d'
r = requests.get(BASE_URL % start)
for image_info in json.loads(r.text)['responseData']['results']:
url = image_info['unescapedUrl']
try:
image_r = requests.get(url)
except ConnectionError, e:
print 'could not download %s' % url
continue
# Remove file-system path characters from name.
title = image_info['titleNoFormatting'].replace('/', '').replace(
'\\', '')
file = open(os.path.join(BASE_PATH, '%s.jpg') % query, 'w')
try:
Image.open(StringIO(image_r.content)).save(file)
file.close()
img = cv2.imread(BASE_PATH + '/' + query + '.jpg')
# save a copy of the image
results = fd.detect_face(img)
if not results:
continue
if float(img.shape[1]) / float(img.shape[0]) > 1.5:
print "Need to crop %s" % query
fx, fy, fw, fh = results['face']
img = img[max(0, fy - 10):min(fy + fh + 10, img.shape[0]),
max(0, fx - 10):min(fx + fw + 10, img.shape[1])]
results = fd.detect_face(img)
if not results:
continue
return (results, img)
except:
# Throw away some gifs...blegh.
print 'could not save %s' % url
continue
finally:
file.close()
start += 4
def run_one_image(input_image,
uv_kpt_ind,
face_ind,
triangles,
uv_coords,
pnet,
rnet,
onet,
x,
y,
Tsess,
ref_texture,
uv_whole_face,
blend_factor=0.35,
minsize=30,
threshold=[0.6, 0.7, 0.7],
factor=0.709,
best_score=0.7,
uv_h=256,
uv_w=256,
image_h=256,
image_w=256):
output_image = input_image.copy()
boxes, pnts = face_detect.detect_face(input_image, minsize, pnet, rnet,
onet, threshold, factor)
faces = process_bbox(boxes, input_image.shape)
is_face = False
for idx, (x0, y1, x1, y0, conf_score) in enumerate(faces):
if conf_score > best_score:
is_face = True
det_face = input_image[int(x0):int(x1), int(y0):int(y1), :]
template_face = det_face.copy()
face_shape = (int(y1) - int(y0), int(x1) - int(x0))
det_face = cv2.resize(det_face, (256, 256)) / 255.
pos = Tsess.run(y, feed_dict={x: det_face[np.newaxis, :, :, :]})
pos = np.squeeze(pos)
max_pos = image_h * 1.1
pos = pos * max_pos
vertices = get_vertices(pos, face_ind, uv_h)
vis_colors = np.ones((vertices.shape[0], 1))
face_mask = cmesh.render.render_texture(vertices.T,
vis_colors.T,
triangles.T,
image_h,
image_w,
c=1)
face_mask = np.squeeze(face_mask > 0).astype(np.float32)
'''
texture = cv2.remap(det_face, pos[:,:,:2].astype(np.float32),
None, interpolation=cv2.INTER_NEAREST,
borderMode=cv2.BORDER_CONSTANT,borderValue=(0))
new_texture = texture*(1. - uv_whole_face[:,:,np.newaxis]) + ref_texture*uv_whole_face[:,:,np.newaxis]
'''
new_texture = ref_texture
new_colors = get_colors_from_texture(new_texture, face_ind, uv_h)
new_image = cmesh.render.render_texture(vertices.T,
new_colors.T,
triangles.T,
image_h,
image_w,
c=3)
new_image = blend_factor * det_face * face_mask[:, :, np.newaxis] + (
1 - blend_factor) * new_image * face_mask[:, :, np.newaxis]
# new_image = color_hist_match(new_image, det_face*face_mask[:,:,np.newaxis])
new_image = det_face * (1. -
face_mask[:, :, np.newaxis]) + new_image
vis_ind = np.argwhere(face_mask > 0)
vis_min = np.min(vis_ind, 0)
vis_max = np.max(vis_ind, 0)
center = (int((vis_min[1] + vis_max[1]) / 2 + 0.5),
int((vis_min[0] + vis_max[0]) / 2 + 0.5))
output = cv2.seamlessClone((new_image * 255.).astype(np.uint8),
(det_face * 255.).astype(np.uint8),
(face_mask * 255.).astype(np.uint8),
center, cv2.NORMAL_CLONE)
temp_ver_image = cv2.resize(output, face_shape)
last_ver_image = image_filter(temp_ver_image, template_face)
output_image[int(x0):int(x1), int(y0):int(y1)] = last_ver_image
return is_face, output_image
def CheckImageFace(self, request, context):
img_str = base64.b64decode(request.imageBuffer)
buffer = BytesIO(img_str)
image = Image.open(buffer)
checked = detect_face(numpy.array(image))
return data_pb2.CheckReply(checkResult=str(checked))
for k,v in faces.items():
if faces[k] == None:
frame = cv2.putText(frame,
"Press {} to load up {} face".format(k[0], k.upper()),
(0,450), cv2.FONT_HERSHEY_COMPLEX,
1, random_color, thickness=2)
break
if not len(preds):
pane = frame[:90, -90:]
pane[:,:] = [0,0,0]
frame[:90, -90:] = pane
else:
face_info = detect_face(preds[0], frame)
#print(face_info)
# DISPLAYTING THE FACE BEING READ
pane = frame[:90, -90:]
pane[:,:] = [0,0,0]
for i in range(3):
for j in range(3):
if (2-j):
pane[i*30 : (i+1)*30, -(3-j)*30:-(2-j)*30] = colors[face_info[i][j]]
else:
pane[i*30 : (i+1)*30, -(3-j)*30:] = colors[face_info[i][j]]
frame[:90, -90:] = pane
def run_one_image(image_path,
uv_kpt_ind,
face_ind,
triangles,
s_uv_coords,
pnet,
rnet,
onet,
x,
y,
Tsess,
minsize=30,
threshold=[0.6, 0.7, 0.7],
factor=0.709,
best_score=0.7,
uv_h=256,
uv_w=256,
image_h=256,
image_w=256):
input_image = cv2.imread(image_path, 1)
output_image = input_image.copy()
boxes, pnts = face_detect.detect_face(input_image, minsize, pnet, rnet,
onet, threshold, factor)
faces = process_bbox(boxes, input_image.shape)
for idx, (x0, y1, x1, y0, conf_score) in enumerate(faces):
if conf_score > best_score:
det_face = input_image[int(x0):int(x1), int(y0):int(y1), :]
face_shape = (int(y1) - int(y0), int(x1) - int(x0))
det_face = cv2.resize(det_face, (256, 256)) / 255.
pos = Tsess.run(y, feed_dict={x: det_face[np.newaxis, :, :, :]})
pos = np.squeeze(pos)
max_pos = image_h
pos = pos * max_pos
vertices = get_vertices(pos, face_ind, uv_h)
from utils.write import write_obj_with_colors
save_vertices = vertices.copy()
save_vertices[:, 1] = image_h - 1 - save_vertices[:, 1]
colors = get_colors(det_face, vertices)
write_obj_with_colors(os.path.join('images', 'test' + '_c.obj'),
save_vertices, triangles, colors)
t_image = (det_face * 255.).astype(np.uint8)
kpt = get_landmarks(pos, uv_kpt_ind)
kpt_origin = plot_kpt(det_face, kpt).astype(np.uint8)
kpt_gray = cv2.cvtColor(kpt_origin, cv2.COLOR_RGB2GRAY)
ret, kpt_mask = cv2.threshold(kpt_gray, 127, 255,
cv2.THRESH_BINARY)
kpt_mask = cv2.bitwise_not(kpt_mask)
kpt_and = cv2.bitwise_and(t_image, t_image, mask=kpt_mask)
kpt_image = cv2.add(kpt_and, kpt_origin)
imsave(os.path.join('images', 'test' + '_kpt.jpg'),
kpt_image / 255.)
t_image = (det_face * 255.).astype(np.uint8)
ver_origin = plot_vertices(det_face, vertices).astype(np.uint8)
ver_gray = cv2.cvtColor(ver_origin, cv2.COLOR_RGB2GRAY)
ret, ver_mask = cv2.threshold(ver_gray, 127, 255,
cv2.THRESH_BINARY)
ver_mask = cv2.bitwise_not(ver_mask)
ver_and = cv2.bitwise_and(t_image, t_image, mask=ver_mask)
ver_image = cv2.add(ver_and, ver_origin)
imsave(os.path.join('images', 'test' + '_ver.jpg'),
ver_image / 255.)
resize_ver_image = cv2.resize(ver_image, face_shape)
output_image[int(x0):int(x1), int(y0):int(y1)] = resize_ver_image
return output_image / 255.
def crop_all(from_root_path, from_dirs, match_names, output_dir):
if not os.path.isdir(output_dir):
os.mkdir(output_dir)
raise RuntimeError('can not find dir %s' % output_dir)
for from_dir in from_dirs:
for root, _, filenames in os.walk(os.path.join(from_root_path, from_dir)):
count = 1
invalid_pts_count = 0
inferenced_from_pts_count=0
total_count = len(filenames) / 2
for filename in filenames:
if filename.endswith('.pts'):
f = filename.split('.')
if not match_names is None and len(match_names)!=0 and not f[0] in match_names:
continue
jpg = os.path.join(root, f[0] + '.jpg')
png = os.path.join(root, f[0] + '.png')
if os.path.isfile(jpg):
url = jpg
elif os.path.isfile(png):
url = png
else:
raise ValueError('unknown file: %s' % os.path.join(root, filename))
path = os.path.join(root, filename)
pts = list(detect.read_pts(path))
def should_reserve(x):
return x != 'data' and x != '..' and x != 'image'
if count % 10 == 0:
print('[%s] %d/%d' % (root, count, total_count))
count += 1
if [1 for pt in pts if pt[0] < 0 or pt[1] < 0]:
# if exists invalid(negative) pt, skip
invalid_pts_count += 1
continue
new_f = '_'.join(filter(should_reserve, os.path.join(root, f[0]).split(os.sep)))
meta_path, img_path, pts_path, = os.path.join(output_dir, new_f + '.meta'), os.path.join(output_dir,
new_f + '.jpg'), os.path.join(
output_dir, new_f + '.pts')
if not os.path.exists(img_path):
# read img file
img = cv2.imread(url)
# filter face
faces = detect.detect_face(img)
if len(faces)==0:
# can not detect face, use face inerenced from pts
inferenced_from_pts_count+=1
face=detect.inference_face_from_pts(img_size=img.shape[:2],pts=pts)
else:
face, pts_num_contained = detect.filter_face(faces, pts)
if pts_num_contained < len(pts) / 2:
# if the detected face does not include most pts, it is invalid, use face inerenced from pts
inferenced_from_pts_count += 1
face = detect.inference_face_from_pts(img_size=img.shape[:2], pts=pts)
zoom_ratio=1.0
while True:
face, can_adjust = detect.adjust_face(img.shape[0:2], face, zoom_ratio)
if not can_adjust:
# adjusted face is out of boundary. use face inerenced from pts
inferenced_from_pts_count += 1
face = detect.inference_face_from_pts(img_size=img.shape[:2], pts=pts)
break
_, pts_num_contained = detect.filter_face([face], pts)
if len(pts) != pts_num_contained:
# some pts is still out of current face, enlarge face
zoom_ratio += 0.1
else:
#current face contains all pts, use it
break
new_img, new_pts = crop(img, face, pts, size=IMG_SIZE)
# write pts file and meta file first
with open(pts_path, 'w') as pts_file:
pts_file.write(print_pts(new_pts))
with open(meta_path, 'w') as meta_file:
src_img_path = url
if src_img_path.startswith('../'):
src_img_path = src_img_path[3:]
# meta file contains two lines
# line 1 : raw image path
# line 2 : face
meta_file.write("%s\n%d,%d,%d,%d" % (src_img_path, *face))
# then write img, thus if img_path does not exists, it is guaranteed to generate pts file
cv2.imwrite(img_path, new_img)
return invalid_pts_count,inferenced_from_pts_count,total_count
def find_character(query):
"""Download full size images from Google image search.
Don't print or republish images without permission.
I used this to train a learning algorithm.
"""
query = query.replace('/', ' ')
path = 'tmp/characters'
BASE_PATH = path
keywords_i = 0
if os.path.exists(BASE_PATH + '/' + query + '.jpg'):
print "Reusing cached image..."
img = cv2.imread(BASE_PATH + '/' + query + '.jpg')
results = False
tries = 0
while not results and tries < 20:
print "Trying to find face again"
results = fd.detect_face(img)
if float(img.shape[1]) / float(img.shape[0]) > 1.5:
print "Need to crop %s" % query
fx, fy, fw, fh = results['face']
img = img[max(0, fy-10):min(fy+fh+10, img.shape[0]), max(0, fx-10):min(fx+fw+10, img.shape[1])]
results = fd.detect_face(img)
if not results:
continue
tries += 1
return (results, img)
if not os.path.exists(BASE_PATH):
os.makedirs(BASE_PATH)
start = 0 # Google's start query string parameter for pagination.
while True:
if start > 3:
keywords_i += 1
start = 0
keyword = ""
character_name = ""
if keywords_i < len(character_lookup_keywords):
keyword = character_lookup_keywords[keywords_i]
character_name = query
elif (keywords_i < 2 * len(character_lookup_keywords)) and query.lower().endswith('s'):
keyword = character_lookup_keywords[keywords_i - len(character_lookup_keywords)]
character_name = query[:-1]
else:
character_name = 'person'
print "Searching for " + character_name + " " + keyword + " " + str(start)
BASE_URL = 'https://ajax.googleapis.com/ajax/services/search/images?'\
'v=1.0&q=' + character_name + '+' + keyword + '&start=%d'
r = requests.get(BASE_URL % start)
for image_info in json.loads(r.text)['responseData']['results']:
url = image_info['unescapedUrl']
try:
image_r = requests.get(url)
except ConnectionError, e:
print 'could not download %s' % url
continue
# Remove file-system path characters from name.
title = image_info['titleNoFormatting'].replace('/', '').replace('\\', '')
file = open(os.path.join(BASE_PATH, '%s.jpg') % query, 'w')
try:
Image.open(StringIO(image_r.content)).save(file)
file.close()
img = cv2.imread(BASE_PATH + '/' + query + '.jpg')
# save a copy of the image
results = fd.detect_face(img)
if not results:
continue
if float(img.shape[1]) / float(img.shape[0]) > 1.5:
print "Need to crop %s" % query
fx, fy, fw, fh = results['face']
img = img[max(0, fy-10):min(fy+fh+10, img.shape[0]), max(0, fx-10):min(fx+fw+10, img.shape[1])]
results = fd.detect_face(img)
if not results:
continue
return (results, img)
except:
# Throw away some gifs...blegh.
print 'could not save %s' % url
continue
finally:
file.close()
start += 4
