def compute_similarities():
t = Timer()
all_descriptors = db.get_all_descriptors()
#print("get_all_descriptors():", t)
print("Faces: %d" % len(all_descriptors), end='')
if len(all_descriptors) < 2:
print()
return
X = Y = np.array([json.loads(f[1]) for f in all_descriptors])
#print("convert to array:", t)
X2 = Y2 = np.sum(np.square(X), axis=-1)
dists = np.sqrt(
np.maximum(X2[:, np.newaxis] + Y2[np.newaxis] - 2 * np.dot(X, Y.T), 0))
#print("calculate dists:", t)
db.delete_similarities()
#print("delete similarities:", t)
num_similarities = 0
for i, j in zip(*np.where(dists < args.similarity_threshold)):
if i != j:
db.insert_similarity(
[all_descriptors[i][0], all_descriptors[j][0], dists[i, j]])
num_similarities += 1
#print("save similarities:", t)
db.commit()
#print("commit:", t)
print(", Similarities: %d, Time: %.2fs" % (num_similarities, t.total()))
def compute_similarities(data_dir, similarity_threshold=0.6, identity_threshold=0.4, criminal_fraction=0.1, **kwargs):
t = Timer()
all_descriptors = db.get_all_descriptors()
descriptors = [json.loads(f[1]) for f in all_descriptors]
face_ids = [f[0] for f in all_descriptors]
num_faces = len(all_descriptors)
#print("get_all_descriptors():", t)
#print("Faces: %d" % len(all_descriptors), end='')
if num_faces < 2:
#print()
return num_faces, 0, 0
X = Y = np.array(descriptors)
#print("convert to array:", t)
X2 = Y2 = np.sum(np.square(X), axis=-1)
dists = np.sqrt(np.maximum(X2[:, np.newaxis] + Y2[np.newaxis] - 2 * np.dot(X, Y.T), 0))
#print("calculate dists:", t)
db.delete_similarities()
#print("delete similarities:", t)
num_similarities = 0
for i, j in zip(*np.where(dists < float(similarity_threshold))):
if i != j:
db.insert_similarity([face_ids[i], face_ids[j], dists[i, j]])
num_similarities += 1
#print("save similarities:", t)
# cluster faces and update labels
descriptors_dlib = [dlib.vector(d) for d in descriptors]
clusters = dlib.chinese_whispers_clustering(descriptors_dlib, float(identity_threshold))
db.update_labels(zip(clusters, face_ids))
num_clusters = len(set(clusters))
if args.save_clusters:
for cluster_num, face_id in zip(clusters, face_ids):
facefile = os.path.realpath(os.path.join(data_dir, args.save_faces, "face_%05d.jpg" % face_id))
clusterdir = os.path.join(data_dir, args.save_clusters, str(cluster_num))
makedirs(clusterdir)
os.symlink(facefile, os.path.join(clusterdir, 'tmpfile'))
os.rename(os.path.join(clusterdir, 'tmpfile'), os.path.join(clusterdir, "face_%05d.jpg" % face_id))
# remove clusters with more than given amount of criminals
criminal_clusters = db.get_clusters_with_criminals(criminal_fraction)
for cluster in criminal_clusters:
db.remove_cluster(cluster['cluster_num'])
db.commit()
#print("commit:", t)
#print(", Similarities: %d, Time: %.2fs" % (num_similarities, t.total()))
return num_faces, num_similarities, num_clusters
def process_image(data_dir, relpath, img, image_type, image_source, frame_num=None, exif_data=None,
gps_lat=None, gps_lon=None, camera_side=None, rotate=0, timestamp=None, **kwargs):
#if int(rotate) != 0:
# assert int(rotate) in [0, 90, 180, 270]
# img = cv2.rotate(img, int(rotate) // 90 - 1)
# print("Rotating", rotate)
image_height, image_width, _ = img.shape
resizepath, resized_height, resized_width = None, image_height, image_width
if args.resize:
img = resize_image(img, args.resize)
resized_height, resized_width, _ = img.shape
if args.save_resized:
filename = os.path.basename(relpath)
resizepath = os.path.join(args.save_resized, filename)
basepath, ext = os.path.splitext(resizepath)
if ext == '' or frame_num is not None:
resizepath = basepath
if frame_num is not None:
resizepath += "_%04d" % frame_num
resizepath += '.jpg'
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = detector(gray, args.upscale)
image_id = db.insert_image([image_type, image_source, relpath, image_width, image_height,
resizepath, resized_width, resized_height, frame_num, exif_data, len(faces),
gps_lat, gps_lon, camera_side, rotate, timestamp])
poses = dlib.full_object_detections()
rects = []
confs = []
for rect in faces:
if args.detector == 'cnn':
confs.append(rect.confidence)
rect = rect.rect
else:
confs.append(1.)
pose = predictor(gray, rect)
poses.append(pose)
rects.append(rect)
# do batched computation of face descriptors
img_rgb = img[...,::-1] # BGR to RGB
descriptors = facerec.compute_face_descriptor(img_rgb, poses, args.jitter)
faceres = []
for i, (rect, conf, pose, descriptor) in enumerate(zip(rects, confs, poses, descriptors)):
face_left = float(rect.left()) / resized_width
face_top = float(rect.top()) / resized_height
face_right = float(rect.right()) / resized_width
face_bottom = float(rect.bottom()) / resized_height
face_width = face_right - face_left
face_height = face_bottom - face_top
landmarks = [[float(p.x) / resized_width, float(p.y) / resized_height] for p in pose.parts()]
descriptor = list(descriptor)
face_id = db.insert_face([image_id, i, face_left, face_top, face_right, face_bottom, face_width, face_height, conf, json.dumps(landmarks), json.dumps(descriptor)])
# draw faces on resized images
if args.draw_faces:
cv2.rectangle(img, (rect.left(), rect.top()), (rect.right(), rect.bottom()), (255, 0, 255), 2)
for p in pose.parts():
cv2.circle(img, (p.x, p.y), 1, color=(0, 255, 255), thickness=-1)
if args.save_faces:
facepath = os.path.join(data_dir, args.save_faces, "face_%05d.jpg" % face_id)
cv2.imwrite(facepath, img[max(rect.top(), 0):rect.bottom(), max(rect.left(), 0):rect.right()])
faceres.append({'face_id': face_id, 'face_num': i, 'left': face_left, 'top': face_top, 'right': face_right, 'bottom': face_bottom, 'width': face_width, 'height': face_height,
'confidence': conf, 'landmarks': landmarks, 'pose_coef': (1 - face_bottom) / face_height})
db.commit()
# save resized image after drawing faces
if args.resize and args.save_resized:
cv2.imwrite(os.path.join(data_dir, resizepath), img)
res = {'relpath': relpath, 'frame_num': frame_num, 'resizepath': resizepath, 'image_id': image_id, 'image_type': image_type, 'num_faces': len(faces), 'faces': faceres}
return len(faces), res
def process_queue():
global images_queue
global grays_queue
global data_queue
global num_images
global num_faces
global num_files
faces_queue = detector(grays_queue,
args.upscale,
batch_size=len(grays_queue))
for faces, img, gray, data in zip(faces_queue, images_queue, grays_queue,
data_queue):
image_id = db.insert_image(data + [len(faces)])
poses = dlib.full_object_detections()
rects = []
for face in faces:
pose = predictor(gray, face.rect)
poses.append(pose)
rects.append(face.rect)
# do batched computation of face descriptors
img_rgb = img[..., ::-1] # BGR to RGB
descriptors = facerec.compute_face_descriptor(img_rgb, poses,
args.jitter)
resized_width = data[5]
resized_height = data[6]
for i, (rect, pose,
descriptor) in enumerate(zip(rects, poses, descriptors)):
face_left = float(rect.left()) / resized_width
face_top = float(rect.top()) / resized_height
face_right = float(rect.right()) / resized_width
face_bottom = float(rect.bottom()) / resized_height
face_width = face_right - face_left
face_height = face_bottom - face_top
landmarks = [[
float(p.x) / resized_width,
float(p.y) / resized_height
] for p in pose.parts()]
descriptor = list(descriptor)
face_id = db.insert_face([
image_id, i, face_left, face_top, face_right, face_bottom,
face_width, face_height,
json.dumps(landmarks),
json.dumps(descriptor)
])
if args.save_faces:
facepath = os.path.join(args.save_faces,
"face_%02d.jpg" % face_id)
cv2.imwrite(
facepath, img[rect.top():rect.bottom(),
rect.left():rect.right()])
num_images += 1
num_faces += len(faces)
db.commit()
images_queue.clear()
grays_queue.clear()
data_queue.clear()
elapsed = time.time() - start_time
print(
"\rFiles: %d, Images: %d, Faces: %d, Elapsed: %.2fs, Images/s: %.1fs" %
(num_files, num_images, num_faces, elapsed, num_images / elapsed),
end='')
def process_image(filepath, img, image_type, frame_num=None, exif_data=None):
image_height, image_width, _ = img.shape
resizepath, resized_height, resized_width = None, image_height, image_width
if args.resize:
img = resize_image(img, args.resize)
if args.save_resized:
filename = os.path.basename(filepath)
resizepath = os.path.join(args.save_resized, filename)
basepath, ext = os.path.splitext(resizepath)
if ext == '' or frame_num is not None:
resizepath = basepath
if frame_num is not None:
resizepath += "_%04d" % frame_num
resizepath += '.jpg'
cv2.imwrite(resizepath, img)
resized_height, resized_width, _ = img.shape
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = detector(gray, args.upscale)
image_id = db.insert_image([
image_type, filepath, image_width, image_height, resizepath,
resized_width, resized_height, frame_num, exif_data,
len(faces)
])
poses = dlib.full_object_detections()
rects = []
for rect in faces:
if args.detector == 'cnn':
rect = rect.rect
pose = predictor(gray, rect)
poses.append(pose)
rects.append(rect)
# do batched computation of face descriptors
img_rgb = img[..., ::-1] # BGR to RGB
descriptors = facerec.compute_face_descriptor(img_rgb, poses, args.jitter)
for i, (rect, pose, descriptor) in enumerate(zip(rects, poses,
descriptors)):
face_left = float(rect.left()) / resized_width
face_top = float(rect.top()) / resized_height
face_right = float(rect.right()) / resized_width
face_bottom = float(rect.bottom()) / resized_height
face_width = face_right - face_left
face_height = face_bottom - face_top
landmarks = [[float(p.x) / resized_width,
float(p.y) / resized_height] for p in pose.parts()]
descriptor = list(descriptor)
face_id = db.insert_face([
image_id, i, face_left, face_top, face_right, face_bottom,
face_width, face_height,
json.dumps(landmarks),
json.dumps(descriptor)
])
if args.save_faces:
facepath = os.path.join(args.save_faces, "face_%02d.jpg" % face_id)
cv2.imwrite(
facepath, img[rect.top():rect.bottom(),
rect.left():rect.right()])
db.commit()
return len(faces)