def chinese_whispers(encodings, threshold=0.5):
"""
Chinese Whispers - an Efficient Graph Clustering Algorithm
and its Application to Natural Language Processing Problems
"""
encodings = [dlib.vector(enc) for enc in encodings]
return dlib.chinese_whispers_clustering(encodings, threshold)
def CLUSTER_TRACKS(DT, threshold):
track_feats = []
for i in DT.keys():
track_feats.append(dlib.vector(DT[i]['BBOX_FEAT'].mean(0)))
CL = defaultdict(dict)
cluster_ids = dlib.chinese_whispers_clustering(track_feats, threshold)
for i in cluster_ids:
try:
CL[i]['BBOX'] = CL[i]['BBOX'] + DT[list(DT.keys())[i]]['BBOX']
CL[i]['Frame_ID'] = CL[i]['Frame_ID'] + DT[list(DT.keys())[i]]['Frame_ID']
CL[i]['BBOX_FEAT'] = CL[i]['BBOX_FEAT'] + DT[list(DT.keys())[i]]['BBOX_FEAT']
CL[i]['ANGLE'] = CL[i]['ANGLE'] + DT[list(DT.keys())[i]]['ANGLE']
CL[i]['IMG'] = CL[i]['IMG'] + DT[list(DT.keys())[i]]['IMG']
CL[i]['AVG_SIZE'] = DT[list(DT.keys())[i]]['AVG_SIZE']
CL[i]['AREA'] = DT[list(DT.keys())[i]]['AREA']
CL[i]['LEN'] = DT[list(DT.keys())[i]]['LEN'] + DT[list(DT.keys())[i]]['LEN']
except:
CL[i]['BBOX'] = DT[list(DT.keys())[i]]['BBOX']
CL[i]['Frame_ID'] = DT[list(DT.keys())[i]]['Frame_ID']
CL[i]['BBOX_FEAT'] = DT[list(DT.keys())[i]]['BBOX_FEAT']
CL[i]['ANGLE'] = DT[list(DT.keys())[i]]['ANGLE']
CL[i]['IMG'] = DT[list(DT.keys())[i]]['IMG']
CL[i]['AVG_SIZE'] = DT[list(DT.keys())[i]]['AVG_SIZE']
CL[i]['AREA'] = DT[list(DT.keys())[i]]['AREA']
CL[i]['LEN'] = DT[list(DT.keys())[i]]['LEN']
return CL
def create_clusters(self, descriptors, d_value=0.5, method="CW"):
print("\nUsing {} cluster method".format(method))
if method == "DB":
#Compute clusters using DBSCAN
clust = DBSCAN(metric="euclidean", n_jobs=1)
clust.fit(descriptors)
clusters = np.unique(clust.labels_)
num_classes = len(np.where(clusters > -1)[0])
clusters = clust.labels_
elif method == "CW":
# Computer clusters using chinese whispers
clusters = dlib.chinese_whispers_clustering(descriptors, d_value)
num_classes = len(set(clusters))
else:
print("Please provide proper method as CW or DB")
exit()
print("Number of clusters: {}".format(num_classes))
print("Clusters: {}".format(clusters))
return clusters
def preprocess_faces(self, faces):
# Cluster the faces with chinese whispers
encodings = [dlib.vector(face['encoding']) for face in faces]
labels = dlib.chinese_whispers_clustering(encodings, 0.5)
selected_faces = []
# Select face most close to average group
groups = list(set(labels))
for group in groups:
# Get indices for each group
indices = [i for i in range(len(labels)) if labels[i] == group]
group_encodings = [faces[i]['encoding'] for i in indices]
# Get centroid for group encodings
avg_group_encoding = np.average(group_encodings, axis=0)
# Get the closest face to the centroid
avg_distance = face_recognition.face_distance(
group_encodings, avg_group_encoding)
min_index = np.argmin(avg_distance)
face_index = indices[min_index]
selected_faces.append(faces[face_index])
return selected_faces
def cluster_faces(self):
self.labels = dlib.chinese_whispers_clustering(self.descriptors, 0.5)
num_classes = len(set(self.labels))
info("Number of clusters: {:,}".format(num_classes))
self.indices = []
for i, label in enumerate(self.labels):
self.indices.append(i)
def clustring(self, faces_info):
for data in faces_info:
encode = data['face_encoding']
self.face_encodings.append(dlib.vector(encode))
labels = dlib.chinese_whispers_clustering(self.face_encodings, 0.5)
labels = np.array(labels)
print("All cluster labels :", labels)
unique_labels = np.unique(labels)
print("Number of unique faces found : ", len(unique_labels))
print("Saving faces..........")
for label in unique_labels:
index = np.where(labels == label)[0]
for i in index:
image_path = self.faces_info[i]['img_path']
image_name = image_path.split('/')[-1].split('.')[0]
image_ext = image_path.split('/')[-1].split('.')[1]
image = cv2.imread(image_path)
output_dir = os.getcwd() + '/' + str(label)
if not os.path.isdir(output_dir):
os.mkdir(str(label))
cv2.imwrite(output_dir + '/' + image_name + '.' + image_ext,
image)
def __clusterize(self, files_faces, debug_out_folder=None):
self.__start_stage(len(files_faces))
encs = []
indexes = list(range(len(files_faces)))
random.shuffle(indexes)
for i in indexes:
for j in range(len(files_faces[i]['faces'])):
encs.append(dlib.vector(
files_faces[i]['faces'][j]['encoding']))
labels = dlib.chinese_whispers_clustering(
encs, self.__threshold_clusterize)
labels = self.__reassign_by_count(labels)
lnum = 0
for i in indexes:
if self.__step_stage():
break
for j in range(len(files_faces[i]['faces'])):
files_faces[i]['faces'][j]['name'] = \
'unknown_{:05d}'.format(labels[lnum])
lnum += 1
if debug_out_folder:
filename = files_faces[i]['filename']
media = tools.load_media(filename,
self.__max_size,
self.__max_video_frames,
self.__video_frames_step)
debug_out_file_name = self.__extract_filename(filename)
self.__save_debug_images(
files_faces[i]['faces'], media,
debug_out_folder, debug_out_file_name)
self.__end_stage()
def cluster(self):
face_labels = dlib.chinese_whispers_clustering(self.all_faces, 0.5)
self.pose_by_label = {}
for i, label in enumerate(face_labels):
if label not in self.pose_by_label:
self.pose_by_label[label] = []
self.pose_by_label[label].append(self.all_poses[i])
def __create_dlib_cluster(config, shape):
data = json.loads(config)
threshod = data.get('threshold') if data.get('threshold') else 0.5
shape = __get_value__(shape)
labels = dlib.chinese_whispers_clustering(shape, threshod)
return len(set(labels)), labels
def cluster_faces(src_dir):
# Load face metadata
faces_df = pd.read_csv(os.path.join(src_dir, 'metadata.csv'))
# Check if clustering already exists
if 'cluster' not in faces_df.columns:
# Chinese whispers clustering
faces_df['embedding'] = faces_df['json_embedding'].apply(json.loads)
X = np.array([x for x in faces_df['embedding']])
faces_df['cluster'] = dlib.chinese_whispers_clustering(
[dlib.vector(x) for x in X], 0.5)
# Persist clustering
faces_df.to_csv(os.path.join(src_dir, 'metadata.csv'), index=False)
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 cluster_embeddings(encodings_path=None):
# Load previously generated embeddings
print("Loading encodings...")
data = pickle.loads(open(Path(encodings_path), "rb").read())
data = np.array(data)
# Specifically grab the encodings from the data array
# If using dlib's Chinese Whispers Clustering, convert to dlib vector format
encodings = [dlib.vector(d["encoding"].squeeze()) for d in data]
# If using KNN, keep in Numpy format
# encodings = [d["encoding"] for d in data]
# encodings = np.asarray(encodings).squeeze()
# Calculate a threshold value for Chinese Whispers
neigh = NearestNeighbors(n_neighbors=5)
nbrs = neigh.fit(encodings)
distances, indices = nbrs.kneighbors(encodings)
distances = np.sort(distances, axis=0)
distances = distances[:, 2]
mean_distance = np.mean(distances)
# plt.plot(distances)
# plt.show()
# Clustering with Chinese Whispers algorithm
labels = dlib.chinese_whispers_clustering(encodings, mean_distance)
# kmeans = KMeans(n_clusters=5, random_state=0).fit(encodings)
# label_ids = np.unique(kmeans.labels_)
# labels = kmeans.labels_
# Determine the total number of unique faces, as well
# as their occurrences
label_ids, counts = np.unique(labels, return_counts=True)
num_unique_faces = len(label_ids)
# Split images into clusters based on labels
image_paths = [d["image_path"] for d in data]
output_folder = image_paths[0].parent.parent.joinpath("clustered_faces")
Path(output_folder).mkdir(parents=True, exist_ok=True)
for i in range(len(image_paths)):
current_label = labels[i]
current_file = image_paths[i]
new_path = output_folder.joinpath(
str(current_label) + "_" + current_file.name)
shutil.copy(current_file, new_path)
def same_person(photo1, photo2):
detector = dlib.get_frontal_face_detector()
sp = dlib.shape_predictor(f"{current_app.config['MODEL_DIR']}/sp.dat")
facerec = dlib.face_recognition_model_v1(f"{current_app.config['MODEL_DIR']}/fr.dat")
p1 = dlib.load_rgb_image(photo1)
p2 = dlib.load_rgb_image(photo2)
face1 = detector(p1, 1)
face2 = detector(p2, 1)
if not face1 or not face2:
return False
for k, d in enumerate(face1):
shape = sp(p1, d)
desc1 = facerec.compute_face_descriptor(p1, shape)
for k, d in enumerate(face2):
shape = sp(p2, d)
desc2 = facerec.compute_face_descriptor(p2, shape)
cv2.rectangle(
p1, (face1[0].left(), face1[0].top()), (face1[0].right(), face1[0].bottom()), (0, 255, 0), 3
)
cv2.rectangle(
p2, (face2[0].left(), face2[0].top()), (face2[0].right(), face2[0].bottom()), (0, 255, 0), 3
)
p1 = cv2.cvtColor(p1, cv2.COLOR_BGR2RGB)
p2 = cv2.cvtColor(p2, cv2.COLOR_BGR2RGB)
cv2.imwrite(photo1, p1)
cv2.imwrite(photo2, p2)
cluster = dlib.chinese_whispers_clustering([desc1, desc2], 0.5)
if len(set(cluster)) == 1:
return True
return False
def cluster():
s = time.time()
query = ''
descriptors = []
dvec = dlib.vectors()
date = input("enter a date in dd-mm-yyy format")
from_time = input("enter start time in hh:mm format")
to_time = input("enter end time in hh:mm format")
data = ptf.retrive(date, from_time, to_time)
for d in data:
descriptors.append(dlib.vector(d))
# Cluster the faces.
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
e = time.time()
print(labels)
print(len(descriptors))
print(len(labels))
labset = set(labels)
print(labset)
num_classes = len(set(labels)) #total number of clusters
print("Number of clusters: {}".format(num_classes))
print(e - s)
return num_classes
def __init__(self, data_path):
names = ['time', 'track']
for i in range(128):
names += ['d{0}'.format(i)]
#
self.data = read_table(data_path, delim_whitespace=True,
header=None, names=names)
self.data.sort_values(by=['track', 'time'], inplace=True)
# create a descriptor list with dlibs descriptor vector
descriptors = []
embeddings = self.data.iloc[:, 2:].values
for each_i in embeddings:
face_descriptor = dlib.vector(each_i)
descriptors.append(face_descriptor)
# returns series of labels [0 0 2 2 2] for each row of embeddings
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
# put the series into a column
self.data['cluster'] = pandas.Series(labels, index=self.data.index)
# TODO: this can be improved by taking highest count of label in each track
# get the label for each track
track_label = self.data.groupby(by='track', as_index=False).first()[
['track', 'cluster']].values
# get unique labels
self.labels = np.unique(track_label[:][:, [1]])
self.starting_point = Annotation(modality='face')
for track, segment in self.data.groupby('track').apply(_to_segment).iteritems():
if not segment:
continue
self.starting_point[segment, track] = track_label[track][1]
def cluster():
labelIdx = 0
encodings = [d["encoding"] for d in helpers.candidate_persons]
labels = dlib.chinese_whispers_clustering(encodings, 0.5)
num_classes = len(set(labels))
for label in labels:
helpers.candidate_persons[labelIdx]["label"] = int(label)
labelIdx += 1
for label in range(num_classes):
face_encs = [
fe for fe in helpers.candidate_persons if fe["label"] == label
]
if (len(face_encs) >= helpers.MIN_FACES_PER_CLUSTER):
mean_enc = np.zeros(128)
for fe in face_encs:
mean_enc += fe["encoding"]
mean_enc = mean_enc / len(face_encs)
helpers.unique_persons.append({
"uuid": uuid.uuid1(),
"Mean": mean_enc
})
uuids = [d["uuid"] for d in helpers.unique_persons]
helpers.candidate_persons = []
def clustering(self,
image_list=None,
output_folder_path=configs_clustering_output_folder,
shape_predictor_path=config_shape_predictor_path,
recognition_model_path=config_recognition_model_path):
"""
人脸聚类函数：使用聚类分析方式进行人脸识别，可以在某一群人中认出特定的人
首先假设最大的群集将包含照片集中的普通人照片
然后提取人脸图像保存150x150格式的最大聚类中
TODO: 可以保存所有大于2的聚类到文件夹中
这里图片不能使用黑白的，否则报错：
RuntimeError: Unsupported image type, must be RGB image.
"""
if not os.path.isdir(output_folder_path):
os.makedirs(output_folder_path)
detector = dlib.get_frontal_face_detector()
shape_predictor = dlib.shape_predictor(shape_predictor_path)
recognition_model = dlib.face_recognition_model_v1(
recognition_model_path)
descriptors = []
images = []
# 找到所有人脸并为每个人脸计算出128维人脸描述器
for i in image_list:
print('正在处理图片: {}'.format(i))
img = io.imread(i)
dets = detector(img, 1)
num_faces = len(dets)
if num_faces == 0:
print("没有找到人脸，文件路径{}".format(i))
continue
print('检测到的人脸数: {}'.format(num_faces))
for k, d in enumerate(dets):
# 得到的人脸特征点/部分在矩形框d中
shape = shape_predictor(img, d)
# 计算128维向量描述的人脸形状
face_descriptor = recognition_model.compute_face_descriptor(
img, shape)
descriptors.append(face_descriptor)
images.append((img, shape))
# 对人脸进行聚类
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
num_classes = len(set(labels))
print("聚类的数量: {}".format(num_classes))
# 找到人脸聚类最多的那个类
biggest_class = None
biggest_class_length = 0
for i in range(0, num_classes):
class_length = len([label for label in labels if label == i])
if class_length > biggest_class_length:
biggest_class_length = class_length
biggest_class = i
print("最大聚类的索引号: {}".format(biggest_class))
print("最大聚类中存储的人脸数: {}".format(biggest_class_length))
# 生成最大聚类生成索引
indices = []
for i, label in enumerate(labels):
if label == biggest_class:
indices.append(i)
print("最大聚类中的图片索引：{}".format(str(indices)))
# 确认输出字典的存在
if not os.path.isdir(output_folder_path):
os.makedirs(output_folder_path)
# 保存提取出来的脸部
print('正在保存最大s聚类到脸部文件夹{}'.format(output_folder_path))
for i, index in enumerate(indices):
img, shape = images[index]
file_path = os.path.join(output_folder_path, 'face_' + str(i))
# 大小(size)和填充(padding)参数默认设置为150x150, 0.25
dlib.save_face_chip(img, shape, file_path, size=150, padding=0.25)
embDim = 128
emb_array = np.zeros((nbFaces,embDim))
faceIx = 0
for file_encs in encs:
for enc_array in file_encs:
dlib_vec = dlib.vector(enc_array)
emb_array[faceIx,:] = enc_array
faceIx = faceIx+1
encodings.append(dlib_vec)
print("[INFO] Clustering faces with Chinese Whispers algorithm")
labels_pred = dlib.chinese_whispers_clustering(encodings, 0.5)
ncols=20
fig = plt.figure(figsize=(20, 4))
thumbnails_labels = sorted(zip(thumbnails,labels_pred), key = lambda t: t[1])
for idx,(npimg,label) in enumerate(thumbnails_labels):
ax = fig.add_subplot(2, ncols, idx+1, xticks=[], yticks=[])
ax.imshow(npimg)
ax.set_title(label)
fig.savefig(
'output_images-thumbnails-classes.png',
bbox_inches='tight',
def cluster(request, eventname):
start = time.time()
md = AzureMediaStorage()
block_blob_service = BlockBlobService(account_name=md.account_name,
account_key=md.account_key)
# Download the pre trained models, unzip them and save them in the save folder as this file
#
predictor_path = 'shape_predictor_5_face_landmarks.dat' #'C:/Users/lenovo/Desktop/PicProcure/events/shape_predictor_5_face_landmarks.dat'
face_rec_model_path = 'dlib_face_recognition_resnet_model_v1.dat'
faces_folder_path = block_blob_service.list_blobs(container_name=eventname)
output_folder = []
check_folder = block_blob_service.list_blobs(container_name='profile-pics')
user_list = Register.objects.all().filter(event_id=Events.objects.get(
event_name=eventname))
username_list = []
for user in user_list:
img = user.user_id.profile_pic
username_list.append(img)
#for f in check_folder:
#username_list.append(f.name)
#print(username_list)
detector = dlib.get_frontal_face_detector() #a detector to find the faces
sp = dlib.shape_predictor(
predictor_path) #shape predictor to find face landmarks
facerec = dlib.face_recognition_model_v1(
face_rec_model_path) #face recognition model
descriptors = []
images = []
output_list = []
for img in check_folder:
print('Processing file:{}', format(img.name))
url = "https://picprocurestorageaccount.blob.core.windows.net/profile-pics/" + img.name
#img1 = dlib.load_rgb_image(urllib.request.urlopen(url).read())
#win = dlib.image_window()
img1 = numpy.array(
Image.open(io.BytesIO(urllib.request.urlopen(url).read())))
#win.set_image(img1)
# Ask the detector to find the bounding boxes of each face. The 1 in the second argument indicates that we should upoutput_listple the image 1 time. This will make everything bigger and allow us to detect more faces.
dets = detector(img1, 1)
print("Number of faces detected: {}".format(len(dets)))
# Now process each face we found.
for k, d in enumerate(dets):
# Get the landmarks/parts for the face in box d.
shape = sp(img1, d)
# Compute the 128D vector that describes the face in img identified by shape.
face_descriptor = facerec.compute_face_descriptor(img1, shape)
descriptors.append(face_descriptor)
images.append(('profile-pics', img.name, img1, shape))
print('profile pics ended')
for f in faces_folder_path:
print("Processing file: {}".format(f.name))
url = "https://picprocurestorageaccount.blob.core.windows.net/" + eventname + '/' + f.name
#img = dlib.load_rgb_image(f)
#win = dlib.image_window()
img = numpy.array(
Image.open(io.BytesIO(urllib.request.urlopen(url).read())))
print('reading completed ' + f.name)
#win.set_image(img)
# Ask the detector to find the bounding boxes of each face. The 1 in the second argument indicates that we should upoutput_listple the image 1 time. This will make everything bigger and allow us to detect more faces.
dets = detector(img, 1)
print("Number of faces detected: {}".format(len(dets)))
# Now process each face we found.
for k, d in enumerate(dets):
# Get the landmarks/parts for the face in box d.
shape = sp(img, d)
# Compute the 128D vector that describes the face in img identified by shape.
face_descriptor = facerec.compute_face_descriptor(img, shape)
descriptors.append(face_descriptor)
images.append((eventname, f.name, img, shape))
print('image appended ' + f.name)
# Cluster the faces.
print("event load completed")
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
num_classes = len(set(labels)) # Total number of clusters
print("Number of clusters: {}".format(num_classes))
for i in range(0, num_classes):
indices = []
class_length = len([label for label in labels if label == i])
for j, label in enumerate(labels):
if label == i:
indices.append(j)
print("Indices of images in the cluster {0} : {1}".format(
str(i), str(indices)))
print("Size of cluster {0} : {1}".format(str(i), str(class_length)))
#output_folder_path = output_folder + '/output' + str(i) # Output folder for each cluster
#os.path.normpath(output_folder_path)
#os.makedirs(output_folder_path)
block_blob_service.create_container(eventname + str(i),
public_access='blob')
# Save each face to the respective cluster folder
print("Saving faces to output folder...")
#img, shape = images[index]
#file_path = os.path.join(output_folder_path,"face_"+str(k)+"_"+str(i))
md.azure_container = eventname + str(i)
output_folder.append(md.azure_container)
for k, index in enumerate(indices):
container, name, img, shape = images[index]
#dlib.save_face_chip(img, shape, file_path, size=1000, padding = 2)
url = "https://picprocurestorageaccount.blob.core.windows.net/" + container + '/' + name
block_blob_service.copy_blob(container_name=md.azure_container,
blob_name=name,
copy_source=url)
# md._save(name,img)
if 0 == k:
output_list.append("ouput/output" + str(i) + "/face_0" + "_" +
str(i) + ".jpg")
for imgs in check_folder:
for output in output_folder:
try:
block_blob_service.get_blob_metadata(container_name=output,
blob_name=imgs.name)
container_name = eventname + '-' + imgs.name.split('.')[0]
block_blob_service.create_container(
container_name=container_name, public_access='blob')
for i in block_blob_service.list_blobs(container_name=output):
url = url = "https://picprocurestorageaccount.blob.core.windows.net/" + output + '/' + i.name
block_blob_service.copy_blob(container_name=container_name,
blob_name=i.name,
copy_source=url)
block_blob_service.delete_container(output)
output_folder.remove(output)
break
except:
pass
block_blob_service.delete_container(eventname)
return HttpResponse("Successfull")
dets = detector(img, 1)
print("Number of faces detected: {}".format(len(dets)))
# Now process each face we found.
for k, d in enumerate(dets):
# Get the landmarks/parts for the face in box d.
shape = sp(img, d)
# Compute the 128D vector that describes the face in img identified by
# shape.
face_descriptor = facerec.compute_face_descriptor(img, shape)
descriptors.append(face_descriptor)
images.append((img, shape))
# Now let's cluster the faces.
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
num_classes = len(set(labels))
print("Number of clusters: {}".format(num_classes))
# Find biggest class
biggest_class = None
biggest_class_length = 0
for i in range(0, num_classes):
class_length = len([label for label in labels if label == i])
if class_length > biggest_class_length:
biggest_class_length = class_length
biggest_class = i
print("Biggest cluster id number: {}".format(biggest_class))
print("Number of faces in biggest cluster: {}".format(biggest_class_length))
def run(self):
if (os.path.exists(self.input_dir) == False):
print(
u'Input directory does not exist.Please check your input directory.'
)
return
if (self.output_dir is None):
father_path = os.path.abspath(
os.path.dirname(self.input_dir) + os.path.sep + ".")
self.output_dir = os.path.join(father_path,
'face_clustering_output')
print(self.output_dir)
# 为后面操作方便，建了几个列表
descriptors = []
images = []
self.log.append(u'----' * 30) # 分界符
# 遍历faces文件夹中所有的图片
for f in glob.glob(os.path.join(self.input_dir, "*.jpg")):
print('Processing file：{}'.format(f))
self.log.append(u'Processing file：{}'.format(f))
# 读取图片
img = self.cv_imread(f)
# 转换到rgb颜色空间
img2 = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
# 检测人脸
dets = self.detector(img2, 1)
print("Number of faces detected: {}".format(len(dets)))
self.log.append(u"Number of faces detected: {}".format(len(dets)))
# 遍历所有的人脸
for index, face in enumerate(dets):
# 检测人脸特征点
shape = self.shape_predictor(img2, face)
# 投影到128D
face_descriptor = self.face_recognizer.compute_face_descriptor(
img2, shape)
# 保存相关信息
descriptors.append(face_descriptor)
images.append((img2, shape))
self.log.append(u'----' * 30)
# 聚类
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
print("labels: {}".format(labels))
self.log.append(u"labels: {}".format(labels))
num_classes = len(set(labels))
print("Number of clusters: {}".format(num_classes))
self.log.append(u"Number of clusters: {}".format(num_classes))
self.log.append(u'----' * 30)
# 为了方便操作，用字典类型保存
face_dict = {}
for i in range(num_classes):
face_dict[i] = []
# print face_dict
for i in range(len(labels)):
face_dict[labels[i]].append(images[i])
# 遍历字典，保存结果
for key in face_dict.keys():
file_dir = os.path.join(self.output_dir, str(key))
if not os.path.isdir(file_dir):
os.makedirs(file_dir)
for index, (image, shape) in enumerate(face_dict[key]):
file_path = os.path.join(file_dir,
'face_' + str(index) + '.jpg')
print file_path
# dlib.save_face_chip(image, shape, file_path, size=150, padding=0.25)
res = dlib.get_face_chip(image, shape, size=150, padding=0.25)
res = cv2.cvtColor(res, cv2.COLOR_RGB2BGR)
cv2.imwrite(file_path, res)
facerec = dlib.face_recognition_model_v1(face_rec_model_path)
paths = glob.glob('faces/*.jpg')
vectors = []
images = []
for path in paths:
img = imread(path)
dets = detector(img, 1)
for i, d in enumerate(dets):
shape = predictor(img, d)
face_vector = facerec.compute_face_descriptor(img, shape)
vectors.append(face_vector)
images.append((img, shape))
# 聚类函数
labels = dlib.chinese_whispers_clustering(vectors, 0.5)
num_classes = len(set(labels))
print('共聚为 %d 类' % num_classes)
biggest_class = Counter(labels).most_common(1)
print(biggest_class)
output_dir = 'most_common'
if not os.path.exists(output_dir):
os.mkdir(output_dir)
face_id = 1
for i in range(len(images)):
if labels[i] == biggest_class[0][0]:
img, shape = images[i]
# 把人脸切出来
dlib.save_face_chip(img,
shape,
def get_feature(video_path, save_path, save_list):
video_files = sorted(os.listdir(video_path))
features = []
idxs = []
faces = []
videos = []
face_attrs = []
fout = open(save_list,'w')
for video_file in video_files:
full_video_path = os.path.join(video_path, video_file)
video = cv2.VideoCapture(full_video_path)
fps = int(video.get(cv2.CAP_PROP_FPS))
num_frame = int(video.get(cv2.CAP_PROP_FRAME_COUNT))
print("processing video_file: {}, num_frame: {}, fps: {}".format(video_file, num_frame, fps))
idx = 0
for i in range(0,num_frame, fps):
video.set(1,i)
ret, frame = video.read()
if not ret:
print("done")
break;
transport.open()
req = ImageReq()
req.name =video_file+"#"+str(i)
req.image_data = cv2.imencode(".jpg",frame)[1]
rsp = client.predict_image(req)
if rsp.status == "OK":
for _, face_feature in enumerate(rsp.face_features):
r = face_feature.region
meta =[video_file, i]
meta.append([int(r.x1),int(r.x2), int(r.y1), int(r.y2)])
meta.append(int(face_feature.age))
meta.append(face_feature.gender)
meta.append(int(face_feature.attractive))
exp_idx = np.argmax(face_feature.exps)
exp = exps[exp_idx]
meta.append(exp)
if (r.x2-r.x1)*(r.y2-r.y1) < MinFaceSize*MinFaceSize:
continue
features.append(dlib.dlib.vector(face_feature.feature))
idxs.append(i)
faces.append(copy.deepcopy(frame[r.y1:r.y2,r.x1:r.x2,:]))
videos.append(video_file)
face_attrs.append(meta)
else:
print("error status: {}".format(rsp.status))
transport.close();
idx+=1
# TODO(crw): for large features size, this would be very slow
# for my case, it cost near 6 hours to complete.
labels = dlib.chinese_whispers_clustering(features, FaceClustingThreshold)
counter = Counter(labels)
ids = []
for id_, cnt in counter.most_common():
if cnt > MinFaceCount:
ids.append(id_)
for i in range(len(labels)):
if labels[i] in ids:
id_path = os.path.join(save_path, str(labels[i]))
if not os.path.exists(id_path):
os.makedirs(id_path)
cv2.imwrite(os.path.join(id_path, videos[i]+"_"+str(idxs[i])+".jpg"), faces[i])
fout.write(str(labels[i])+"\t"+str(face_attrs[i])+"\n")
fout.close()
def Get_face_clustered_labels(faces_folder_path):
import sys
import os
import dlib
import glob
from tqdm import tqdm
#print (faces_folder_path, os.listdir(faces_folder_path))
# Download the pre trained models, unzip them and save them in the save folder as this file
predictor_path = 'FaceClust/shape_predictor_5_face_landmarks.dat' # Download from http://dlib.net/files/dlib_face_recognition_resnet_model_v1.dat.bz2
face_rec_model_path = 'FaceClust/dlib_face_recognition_resnet_model_v1.dat' # Download from http://dlib.net/files/shape_predictor_5_face_landmarks.dat.bz2
detector = dlib.cnn_face_detection_model_v1(
'FaceClust/mmod_human_face_detector.dat'
) #a detector to find the faces
sp = dlib.shape_predictor(
predictor_path) #shape predictor to find face landmarks
facerec = dlib.face_recognition_model_v1(
face_rec_model_path) #face recognition model
descriptors = []
images = []
# Load the images from input folder
FACE_PATHS = []
dn = glob.glob(os.path.join(faces_folder_path, "*.jpg"))
f = open('/tmp/ProgressN', 'w+')
f.write(str(len(dn)))
f.close()
ccc = 1
for f in tqdm(dn):
ff = open('/tmp/ProgressI', 'w+')
ff.write(str(ccc))
ff.close()
ccc = ccc + 1
#print("Processing file: {}".format(f))
img = dlib.load_rgb_image(f)
# Ask the detector to find the bounding boxes of each face. The 1 in the second argument indicates that we should upsample the image 1 time. This will make everything bigger and allow us to detect more faces.
dets = detector(img, 1)
#print("Number of faces detected: {}".format(len(dets)))
# Now process each face we found.
for k, d in enumerate(dets):
# Get the landmarks/parts for the face in box d.
shape = sp(img, d.rect)
# Compute the 128D vector that describes the face in img identified by shape.
face_descriptor = facerec.compute_face_descriptor(img, shape)
descriptors.append(face_descriptor)
FACE_PATHS.append(f)
# Cluster the faces.
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
num_classes = len(set(labels)) # Total number of clusters
print("Number of clusters: {}".format(num_classes))
face_label_dicts = {}
#print ('xxxxxx')
#print (FACE_PATHS)
for i, j in zip(FACE_PATHS, labels):
if j in face_label_dicts:
face_label_dicts[j].append(i)
else:
face_label_dicts[j] = [i]
#print ('xxxxxx')
if os.path.isfile('/tmp/ProgressN'): os.remove('/tmp/ProgressN')
if os.path.isfile('/tmp/ProgressI'): os.remove('/tmp/ProgressI')
del detector
del sp
del facerec
del dlib
return face_label_dicts
def __cluster(faces_vectors: list, threshold=0.5):
return dlib.chinese_whispers_clustering(faces_vectors, threshold)
def cluster_faces(descriptors, threshold=0.45):
return dlib.chinese_whispers_clustering(descriptors, threshold)
def process():
mylogger = loger.getLoger("Whisper", Constants.boltpath + "logs")
try:
timenow = Blocker.current_milli_time()
# Generic models
video = DatabaseSession.session.query(Videos).first()
# Some paths
detector = dlib.get_frontal_face_detector()
sp = dlib.shape_predictor(
Constants.predictor_path) # 128D face descriptor predictor
facerec = dlib.face_recognition_model_v1(Constants.face_rec_model_path)
descriptors = []
images = []
# Now find all the faces and compute 128D face descriptors for each face.
for f in glob.glob(os.path.join(Constants.faces_folder_path, "*.png")):
mylogger.info("Processing file: {}".format(f))
img = dlib.load_rgb_image(f)
# Ask the detector to find the bounding boxes of each face. The 1 in the
# second argument indicates that we should upsample the image 1 time. This
# will make everything bigger and allow us to detect more faces.
dets = detector(img, 1)
mylogger.info("Number of faces detected: {}".format(len(dets)))
# Now process each face we found.
for k, d in enumerate(dets):
# Get the landmarks/parts for the face in box d.
shape = sp(img, d)
# Compute the 128D vector that describes the face in img identified by
# shape.
face_descriptor = facerec.compute_face_descriptor(img, shape)
descriptors.append(face_descriptor)
images.append((img, shape))
# Now let's cluster the faces.
labels = dlib.chinese_whispers_clustering(descriptors,
Constants.TOLERANCE_WHISPER)
num_classes = len(set(labels))
mylogger.info("Number of clusters: {}".format(num_classes))
if num_classes > 0:
# Find the indices for the biggest class
indices = []
counts = 0
DatabaseSession.session.query(Wishper).filter(
Wishper.video == video).delete()
DatabaseSession.session.commit()
for i, label in enumerate(labels):
# Ensure output directory exists
output_folder_path_real = Constants.output_folder_path + str(
label)
if not os.path.isdir(output_folder_path_real):
os.makedirs(output_folder_path_real)
# Save the extracted faces
mylogger.info("Saving faces cluster to output folder...")
img, shape = images[i]
file_path = os.path.join(output_folder_path_real,
"face_" + str(counts))
# The size and padding arguments are optional size=300x300 and padding=0.25
dlib.save_face_chip(img,
shape,
file_path,
size=300,
padding=0.25)
counts = counts + 1
whisper = Wishper(alias=file_path,
original_image=file_path,
video=video,
group=str(label),
path=file_path + ".jpg")
DatabaseSession.session.add(whisper)
DatabaseSession.session.commit()
except Exception as inst:
mylogger.error("Python error.")
mylogger.error(type(inst))
mylogger.error(inst)
def cluster_faces_by_CW(data, threshold=0.5):
encodings = [dlib.vector(d["encoding"]) for d in data]
labels = dlib.chinese_whispers_clustering(encodings, threshold)
return labels
def cluster_faces_in_class(args):
preds_per_person = utils.load_faces_from_csv(args.db, args.imgs_root)
if preds_per_person.get(args.cls) == None:
print('Class {} not found.'.format(args.cls))
return
descriptors = []
for i, p in enumerate(preds_per_person[args.cls]):
descriptors.append(dlib.vector(p[2]))
# cluster the faces
print('clustering...')
all_indices = []
all_lengths = []
if 1:
# chinese whispers
labels = dlib.chinese_whispers_clustering(descriptors, args.threshold)
num_classes = len(set(labels))
print("Number of clusters: {}".format(num_classes))
for j in range(0, num_classes):
class_length = len([label for label in labels if label == j])
if class_length >= args.min_members:
indices = []
for i, label in enumerate(labels):
if label == j:
indices.append(i)
all_indices.append(indices)
all_lengths.append(class_length)
else:
#DBSCAN
from sklearn.cluster import DBSCAN
clt = DBSCAN(eps=args.threshold,
metric="euclidean",
n_jobs=4,
min_samples=args.min_members)
clt.fit(descriptors)
labels = np.unique(clt.labels_)
num_classes = len(np.where(labels > -1)[0])
if num_classes > 1: # to be checked!!
print("Number of clusters: {}".format(num_classes))
for j in labels:
idxs = np.where(clt.labels_ == j)[0]
class_length = len(idxs)
indices = []
for i in idxs:
indices.append(i)
all_indices.append(indices)
all_lengths.append(class_length)
sort_index = np.argsort(np.array(all_lengths))[::-1]
# Move the clustered faces to individual groups
print('Moving the clustered faces to the database.')
to_delete = []
for i in sort_index[:args.max_clusters]:
cluster_name = "group_" + str(i)
# export to folders
if args.export:
cluster_path = os.path.join(args.outdir, cluster_name)
if not os.path.isdir(cluster_path):
os.makedirs(cluster_path)
to_delete += all_indices[i]
for n, index in enumerate(all_indices[i]):
utils.insert_element_preds_per_person(preds_per_person, args.cls,
index, cluster_name)
if args.export:
file_name = os.path.join(
cluster_path,
'face_' + cluster_name + '_' + str(n) + '.jpg')
utils.save_face_crop(file_name,
preds_per_person[args.cls][index][1],
preds_per_person[args.cls][index][0][1])
to_delete = sorted(to_delete)
to_delete.reverse()
for i in to_delete:
preds_per_person[cls].pop(
i
) # if not, they would exist double, in 'deleted' and in the cluster group
# utils.delete_element_preds_per_person(preds_per_person, args.cls, i)
utils.export_persons_to_csv(preds_per_person, args.imgs_root, args.db)
for k, d in enumerate(dets):
# Get the landmarks/parts for the face in box d.
shape = sp(img, d)
# Compute the 128D vector that describes the face in img identified by shape.
face_descriptor = facerec.compute_face_descriptor(img, shape)
descriptors.append(face_descriptor)
images.append((img, shape))
except:
continue
cv2.imshow('frame', frame)
if cv2.waitKey(20) & 0xFF == ord('q'):
break
cap.release()
# Cluster the faces.
labels = dlib.chinese_whispers_clustering(descriptors, 0.5)
num_classes = len(set(labels)) # Total number of clusters
print("Number of clusters: {}".format(num_classes))
for i in range(0, num_classes):
indices = []
class_length = len([label for label in labels if label == i])
for j, label in enumerate(labels):
if label == i:
indices.append(j)
print("Indices of images in the cluster {0} : {1}".format(str(i),str(indices)))
print("Size of cluster {0} : {1}".format(str(i),str(class_length)))
output_folder_path = output_folder +'./'+ str(i) # Output folder for each cluster
os.path.normpath(output_folder_path)
os.makedirs(output_folder_path)
files = glob.glob(path.join(PICKLES_DIR, '*.pickle'))
print(len(files))
for picklefile in files:
d = pickle.loads(open(picklefile, "rb").read())
data.extend(d)
#data = pickle.loads(open(args["encodings"], "rb").read())
#data = np.array(data)
print(len(data))
encodings = [dlib.vector(d["encoding"]) for d in data]
#[start:end]]
print(len(encodings))
#print(type(encodings))
#print(encodings[0])
# cluster the embeddings
print("[INFO] clustering...")
#clt = DBSCAN(metric="euclidean", n_jobs=-1)
#clt.fit(encodings)
labels = dlib.chinese_whispers_clustering(encodings, THRESHOLD)
tup = (data, labels)
pickleFile = 'clusters-%.3f.pickle' % THRESHOLD
with open(pickleFile, "wb") as f:
f.write(pickle.dumps(tup))
