github/deepface
1
1
Fork 0
mirror of https://github.com/serengil/deepface.git synced 2025-07-24 19:00:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'serengil:master' into master

Browse Source
This commit is contained in:
Tyas 2024-10-10 16:50:58 +07:00 committed by GitHub
commit 49c041945a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
25 changed files with 361 additions and 211 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
deepface/DeepFace.py
View File

@ -364,7 +364,7 @@ def find(
silent=silent,
refresh_database=refresh_database,
anti_spoofing=anti_spoofing,
batched=batched
batched=batched,
)

98
deepface/commons/weight_utils.py
View File

@ -11,6 +11,7 @@ import gdown
from deepface.commons import folder_utils, package_utils
from deepface.commons.logger import Logger
tf_version = package_utils.get_tf_major_version()
if tf_version == 1:
from keras.models import Sequential
@ -19,6 +20,8 @@ else:
logger = Logger()
# pylint: disable=line-too-long, use-maxsplit-arg
ALLOWED_COMPRESS_TYPES = ["zip", "bz2"]
@ -95,3 +98,98 @@ def load_model_weights(model: Sequential, weight_file: str) -> Sequential:
"and copying it to the target folder."
) from err
return model
def download_all_models_in_one_shot() -> None:
"""
Download all model weights in one shot
"""
# import model weights from module here to avoid circular import issue
from deepface.models.facial_recognition.VGGFace import WEIGHTS_URL as VGGFACE_WEIGHTS
from deepface.models.facial_recognition.Facenet import FACENET128_WEIGHTS, FACENET512_WEIGHTS
from deepface.models.facial_recognition.OpenFace import WEIGHTS_URL as OPENFACE_WEIGHTS
from deepface.models.facial_recognition.FbDeepFace import WEIGHTS_URL as FBDEEPFACE_WEIGHTS
from deepface.models.facial_recognition.ArcFace import WEIGHTS_URL as ARCFACE_WEIGHTS
from deepface.models.facial_recognition.DeepID import WEIGHTS_URL as DEEPID_WEIGHTS
from deepface.models.facial_recognition.SFace import WEIGHTS_URL as SFACE_WEIGHTS
from deepface.models.facial_recognition.GhostFaceNet import WEIGHTS_URL as GHOSTFACENET_WEIGHTS
from deepface.models.facial_recognition.Dlib import WEIGHT_URL as DLIB_FR_WEIGHTS
from deepface.models.demography.Age import WEIGHTS_URL as AGE_WEIGHTS
from deepface.models.demography.Gender import WEIGHTS_URL as GENDER_WEIGHTS
from deepface.models.demography.Race import WEIGHTS_URL as RACE_WEIGHTS
from deepface.models.demography.Emotion import WEIGHTS_URL as EMOTION_WEIGHTS
from deepface.models.spoofing.FasNet import (
FIRST_WEIGHTS_URL as FASNET_1ST_WEIGHTS,
SECOND_WEIGHTS_URL as FASNET_2ND_WEIGHTS,
)
from deepface.models.face_detection.Ssd import (
MODEL_URL as SSD_MODEL,
WEIGHTS_URL as SSD_WEIGHTS,
)
from deepface.models.face_detection.Yolo import (
WEIGHT_URL as YOLOV8_WEIGHTS,
WEIGHT_NAME as YOLOV8_WEIGHT_NAME,
)
from deepface.models.face_detection.YuNet import WEIGHTS_URL as YUNET_WEIGHTS
from deepface.models.face_detection.Dlib import WEIGHTS_URL as DLIB_FD_WEIGHTS
from deepface.models.face_detection.CenterFace import WEIGHTS_URL as CENTERFACE_WEIGHTS
WEIGHTS = [
# facial recognition
VGGFACE_WEIGHTS,
FACENET128_WEIGHTS,
FACENET512_WEIGHTS,
OPENFACE_WEIGHTS,
FBDEEPFACE_WEIGHTS,
ARCFACE_WEIGHTS,
DEEPID_WEIGHTS,
SFACE_WEIGHTS,
{
"filename": "ghostfacenet_v1.h5",
"url": GHOSTFACENET_WEIGHTS,
},
DLIB_FR_WEIGHTS,
# demography
AGE_WEIGHTS,
GENDER_WEIGHTS,
RACE_WEIGHTS,
EMOTION_WEIGHTS,
# spoofing
FASNET_1ST_WEIGHTS,
FASNET_2ND_WEIGHTS,
# face detection
SSD_MODEL,
SSD_WEIGHTS,
{
"filename": YOLOV8_WEIGHT_NAME,
"url": YOLOV8_WEIGHTS,
},
YUNET_WEIGHTS,
DLIB_FD_WEIGHTS,
CENTERFACE_WEIGHTS,
]
for i in WEIGHTS:
if isinstance(i, str):
url = i
filename = i.split("/")[-1]
compress_type = None
# if compressed file will be downloaded, get rid of its extension
if filename.endswith(tuple(ALLOWED_COMPRESS_TYPES)):
for ext in ALLOWED_COMPRESS_TYPES:
compress_type = ext
if filename.endswith(f".{ext}"):
filename = filename[: -(len(ext) + 1)]
break
elif isinstance(i, dict):
filename = i["filename"]
url = i["url"]
else:
raise ValueError("unimplemented scenario")
logger.info(
f"Downloading {url} to ~/.deepface/weights/{filename} with {compress_type} compression"
)
download_weights_if_necessary(
file_name=filename, source_url=url, compress_type=compress_type
)

9
deepface/models/Detector.py
View File

@ -6,7 +6,7 @@ import numpy as np
# Notice that all facial detector models must be inherited from this class
# pylint: disable=unnecessary-pass, too-few-public-methods
# pylint: disable=unnecessary-pass, too-few-public-methods, too-many-instance-attributes
class Detector(ABC):
@abstractmethod
def detect_faces(self, img: np.ndarray) -> List["FacialAreaRegion"]:
@ -45,6 +45,7 @@ class FacialAreaRegion:
confidence (float, optional): Confidence score associated with the face detection.
Default is None.
"""
x: int
y: int
w: int
@ -52,6 +53,9 @@ class FacialAreaRegion:
left_eye: Optional[Tuple[int, int]] = None
right_eye: Optional[Tuple[int, int]] = None
confidence: Optional[float] = None
nose: Optional[Tuple[int, int]] = None
mouth_right: Optional[Tuple[int, int]] = None
mouth_left: Optional[Tuple[int, int]] = None
@dataclass
@ -63,7 +67,8 @@ class DetectedFace:
img (np.ndarray): detected face image as numpy array
facial_area (FacialAreaRegion): detected face's metadata (e.g. bounding box)
confidence (float): confidence score for face detection
"""
"""
img: np.ndarray
facial_area: FacialAreaRegion
confidence: float

11
deepface/models/demography/Age.py
View File

@ -23,6 +23,10 @@ else:
# ----------------------------------------
WEIGHTS_URL = (
"https://github.com/serengil/deepface_models/releases/download/v1.0/age_model_weights.h5"
)
# pylint: disable=too-few-public-methods
class ApparentAgeClient(Demography):
"""
@ -41,7 +45,7 @@ class ApparentAgeClient(Demography):
def load_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/age_model_weights.h5",
url=WEIGHTS_URL,
) -> Model:
"""
Construct age model, download its weights and load
@ -70,12 +74,11 @@ def load_model(
file_name="age_model_weights.h5", source_url=url
)
age_model = weight_utils.load_model_weights(
model=age_model, weight_file=weight_file
)
age_model = weight_utils.load_model_weights(model=age_model, weight_file=weight_file)
return age_model
def find_apparent_age(age_predictions: np.ndarray) -> np.float64:
"""
Find apparent age prediction from a given probas of ages

20
deepface/models/demography/Emotion.py
View File

@ -7,11 +7,6 @@ from deepface.commons import package_utils, weight_utils
from deepface.models.Demography import Demography
from deepface.commons.logger import Logger
logger = Logger()
# -------------------------------------------
# pylint: disable=line-too-long
# -------------------------------------------
# dependency configuration
tf_version = package_utils.get_tf_major_version()
@ -28,12 +23,17 @@ else:
Dense,
Dropout,
)
# -------------------------------------------
# Labels for the emotions that can be detected by the model.
labels = ["angry", "disgust", "fear", "happy", "sad", "surprise", "neutral"]
# pylint: disable=too-few-public-methods
logger = Logger()
# pylint: disable=line-too-long, disable=too-few-public-methods
WEIGHTS_URL = "https://github.com/serengil/deepface_models/releases/download/v1.0/facial_expression_model_weights.h5"
class EmotionClient(Demography):
"""
Emotion model class
@ -56,7 +56,7 @@ class EmotionClient(Demography):
def load_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/facial_expression_model_weights.h5",
url=WEIGHTS_URL,
) -> Sequential:
"""
Consruct emotion model, download and load weights
@ -96,8 +96,6 @@ def load_model(
file_name="facial_expression_model_weights.h5", source_url=url
)
model = weight_utils.load_model_weights(
model=model, weight_file=weight_file
)
model = weight_utils.load_model_weights(model=model, weight_file=weight_file)
return model

5
deepface/models/demography/Gender.py
View File

@ -21,7 +21,8 @@ if tf_version == 1:
else:
from tensorflow.keras.models import Model, Sequential
from tensorflow.keras.layers import Convolution2D, Flatten, Activation
# -------------------------------------
WEIGHTS_URL="https://github.com/serengil/deepface_models/releases/download/v1.0/gender_model_weights.h5"
# Labels for the genders that can be detected by the model.
labels = ["Woman", "Man"]
@ -43,7 +44,7 @@ class GenderClient(Demography):
def load_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/gender_model_weights.h5",
url=WEIGHTS_URL,
) -> Model:
"""
Construct gender model, download its weights and load

18
deepface/models/demography/Race.py
View File

@ -7,11 +7,8 @@ from deepface.commons import package_utils, weight_utils
from deepface.models.Demography import Demography
from deepface.commons.logger import Logger
logger = Logger()
# --------------------------
# pylint: disable=line-too-long
# --------------------------
# dependency configurations
tf_version = package_utils.get_tf_major_version()
@ -21,10 +18,15 @@ if tf_version == 1:
else:
from tensorflow.keras.models import Model, Sequential
from tensorflow.keras.layers import Convolution2D, Flatten, Activation
# --------------------------
WEIGHTS_URL = (
"https://github.com/serengil/deepface_models/releases/download/v1.0/race_model_single_batch.h5"
)
# Labels for the ethnic phenotypes that can be detected by the model.
labels = ["asian", "indian", "black", "white", "middle eastern", "latino hispanic"]
logger = Logger()
# pylint: disable=too-few-public-methods
class RaceClient(Demography):
"""
@ -42,7 +44,7 @@ class RaceClient(Demography):
def load_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/race_model_single_batch.h5",
url=WEIGHTS_URL,
) -> Model:
"""
Construct race model, download its weights and load
@ -69,8 +71,6 @@ def load_model(
file_name="race_model_single_batch.h5", source_url=url
)
race_model = weight_utils.load_model_weights(
model=race_model, weight_file=weight_file
)
race_model = weight_utils.load_model_weights(model=race_model, weight_file=weight_file)
return race_model

3
deepface/models/face_detection/Dlib.py
View File

@ -11,6 +11,7 @@ from deepface.commons.logger import Logger
logger = Logger()
WEIGHTS_URL="http://dlib.net/files/shape_predictor_5_face_landmarks.dat.bz2"
class DlibClient(Detector):
def __init__(self):
@ -34,7 +35,7 @@ class DlibClient(Detector):
# check required file exists in the home/.deepface/weights folder
weight_file = weight_utils.download_weights_if_necessary(
file_name="shape_predictor_5_face_landmarks.dat",
source_url="http://dlib.net/files/shape_predictor_5_face_landmarks.dat.bz2",
source_url=WEIGHTS_URL,
compress_type="bz2",
)

12
deepface/models/face_detection/RetinaFace.py
View File

@ -42,10 +42,19 @@ class RetinaFaceClient(Detector):
# retinaface sets left and right eyes with respect to the person
left_eye = identity["landmarks"]["left_eye"]
right_eye = identity["landmarks"]["right_eye"]
nose = identity["landmarks"].get("nose")
mouth_right = identity["landmarks"].get("mouth_right")
mouth_left = identity["landmarks"].get("mouth_left")
# eyes are list of float, need to cast them tuple of int
left_eye = tuple(int(i) for i in left_eye)
right_eye = tuple(int(i) for i in right_eye)
if nose is not None:
nose = tuple(int(i) for i in nose)
if mouth_right is not None:
mouth_right = tuple(int(i) for i in mouth_right)
if mouth_left is not None:
mouth_left = tuple(int(i) for i in mouth_left)
confidence = identity["score"]
@ -57,6 +66,9 @@ class RetinaFaceClient(Detector):
left_eye=left_eye,
right_eye=right_eye,
confidence=confidence,
nose=nose,
mouth_left=mouth_left,
mouth_right=mouth_right,
)
resp.append(facial_area)

7
deepface/models/face_detection/Ssd.py
View File

@ -16,6 +16,9 @@ logger = Logger()
# pylint: disable=line-too-long, c-extension-no-member
MODEL_URL = "https://github.com/opencv/opencv/raw/3.4.0/samples/dnn/face_detector/deploy.prototxt"
WEIGHTS_URL = "https://github.com/opencv/opencv_3rdparty/raw/dnn_samples_face_detector_20170830/res10_300x300_ssd_iter_140000.caffemodel"
class SsdClient(Detector):
def __init__(self):
@ -31,13 +34,13 @@ class SsdClient(Detector):
# model structure
output_model = weight_utils.download_weights_if_necessary(
file_name="deploy.prototxt",
source_url="https://github.com/opencv/opencv/raw/3.4.0/samples/dnn/face_detector/deploy.prototxt",
source_url=MODEL_URL,
)
# pre-trained weights
output_weights = weight_utils.download_weights_if_necessary(
file_name="res10_300x300_ssd_iter_140000.caffemodel",
source_url="https://github.com/opencv/opencv_3rdparty/raw/dnn_samples_face_detector_20170830/res10_300x300_ssd_iter_140000.caffemodel",
source_url=WEIGHTS_URL,
)
try:

4
deepface/models/face_detection/Yolo.py
View File

@ -12,7 +12,7 @@ from deepface.commons.logger import Logger
logger = Logger()
# Model's weights paths
PATH = ".deepface/weights/yolov8n-face.pt"
WEIGHT_NAME = "yolov8n-face.pt"
# Google Drive URL from repo (https://github.com/derronqi/yolov8-face) ~6MB
WEIGHT_URL = "https://drive.google.com/uc?id=1qcr9DbgsX3ryrz2uU8w4Xm3cOrRywXqb"
@ -39,7 +39,7 @@ class YoloClient(Detector):
) from e
weight_file = weight_utils.download_weights_if_necessary(
file_name="yolov8n-face.pt", source_url=WEIGHT_URL
file_name=WEIGHT_NAME, source_url=WEIGHT_URL
)
# Return face_detector

5
deepface/models/face_detection/YuNet.py
View File

@ -13,6 +13,9 @@ from deepface.commons.logger import Logger
logger = Logger()
# pylint:disable=line-too-long
WEIGHTS_URL = "https://github.com/opencv/opencv_zoo/raw/main/models/face_detection_yunet/face_detection_yunet_2023mar.onnx"
class YuNetClient(Detector):
def __init__(self):
@ -41,7 +44,7 @@ class YuNetClient(Detector):
# pylint: disable=C0301
weight_file = weight_utils.download_weights_if_necessary(
file_name="face_detection_yunet_2023mar.onnx",
source_url="https://github.com/opencv/opencv_zoo/raw/main/models/face_detection_yunet/face_detection_yunet_2023mar.onnx",
source_url=WEIGHTS_URL,
)
try:

4
deepface/models/facial_recognition/ArcFace.py
View File

@ -42,6 +42,8 @@ else:
Dense,
)
WEIGHTS_URL="https://github.com/serengil/deepface_models/releases/download/v1.0/arcface_weights.h5"
# pylint: disable=too-few-public-methods
class ArcFaceClient(FacialRecognition):
"""
@ -56,7 +58,7 @@ class ArcFaceClient(FacialRecognition):
def load_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/arcface_weights.h5",
url=WEIGHTS_URL,
) -> Model:
"""
Construct ArcFace model, download its weights and load

5
deepface/models/facial_recognition/DeepID.py
View File

@ -34,8 +34,7 @@ else:
# pylint: disable=line-too-long
# -------------------------------------
WEIGHTS_URL="https://github.com/serengil/deepface_models/releases/download/v1.0/deepid_keras_weights.h5"
# pylint: disable=too-few-public-methods
class DeepIdClient(FacialRecognition):
@ -51,7 +50,7 @@ class DeepIdClient(FacialRecognition):
def load_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/deepid_keras_weights.h5",
url=WEIGHTS_URL,
) -> Model:
"""
Construct DeepId model, download its weights and load

3
deepface/models/facial_recognition/Dlib.py
View File

@ -12,6 +12,7 @@ from deepface.commons.logger import Logger
logger = Logger()
# pylint: disable=too-few-public-methods
WEIGHT_URL = "http://dlib.net/files/dlib_face_recognition_resnet_model_v1.dat.bz2"
class DlibClient(FacialRecognition):
@ -70,7 +71,7 @@ class DlibResNet:
weight_file = weight_utils.download_weights_if_necessary(
file_name="dlib_face_recognition_resnet_model_v1.dat",
source_url="http://dlib.net/files/dlib_face_recognition_resnet_model_v1.dat.bz2",
source_url=WEIGHT_URL,
compress_type="bz2",
)

20
deepface/models/facial_recognition/Facenet.py
View File

@ -39,6 +39,14 @@ else:
from tensorflow.keras.layers import add
from tensorflow.keras import backend as K
# pylint:disable=line-too-long
FACENET128_WEIGHTS = (
"https://github.com/serengil/deepface_models/releases/download/v1.0/facenet_weights.h5"
)
FACENET512_WEIGHTS = (
"https://github.com/serengil/deepface_models/releases/download/v1.0/facenet512_weights.h5"
)
# --------------------------------
# pylint: disable=too-few-public-methods
@ -1654,7 +1662,7 @@ def InceptionResNetV1(dimension: int = 128) -> Model:
def load_facenet128d_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/facenet_weights.h5",
url=FACENET128_WEIGHTS,
) -> Model:
"""
Construct FaceNet-128d model, download weights and then load weights
@ -1668,15 +1676,13 @@ def load_facenet128d_model(
weight_file = weight_utils.download_weights_if_necessary(
file_name="facenet_weights.h5", source_url=url
)
model = weight_utils.load_model_weights(
model=model, weight_file=weight_file
)
model = weight_utils.load_model_weights(model=model, weight_file=weight_file)
return model
def load_facenet512d_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/facenet512_weights.h5",
url=FACENET512_WEIGHTS,
) -> Model:
"""
Construct FaceNet-512d model, download its weights and load
@ -1689,8 +1695,6 @@ def load_facenet512d_model(
weight_file = weight_utils.download_weights_if_necessary(
file_name="facenet512_weights.h5", source_url=url
)
model = weight_utils.load_model_weights(
model=model, weight_file=weight_file
)
model = weight_utils.load_model_weights(model=model, weight_file=weight_file)
return model

6
deepface/models/facial_recognition/FbDeepFace.py
View File

@ -30,9 +30,9 @@ else:
Dropout,
)
# -------------------------------------
# pylint: disable=line-too-long, too-few-public-methods
WEIGHTS_URL="https://github.com/swghosh/DeepFace/releases/download/weights-vggface2-2d-aligned/VGGFace2_DeepFace_weights_val-0.9034.h5.zip"
class DeepFaceClient(FacialRecognition):
"""
Fb's DeepFace model class
@ -54,7 +54,7 @@ class DeepFaceClient(FacialRecognition):
def load_model(
url="https://github.com/swghosh/DeepFace/releases/download/weights-vggface2-2d-aligned/VGGFace2_DeepFace_weights_val-0.9034.h5.zip",
url=WEIGHTS_URL,
) -> Model:
"""
Construct DeepFace model, download its weights and load

8
deepface/models/facial_recognition/GhostFaceNet.py
View File

@ -48,7 +48,7 @@ else:
# pylint: disable=line-too-long, too-few-public-methods, no-else-return, unsubscriptable-object, comparison-with-callable
PRETRAINED_WEIGHTS = "https://github.com/HamadYA/GhostFaceNets/releases/download/v1.2/GhostFaceNet_W1.3_S1_ArcFace.h5"
WEIGHTS_URL = "https://github.com/HamadYA/GhostFaceNets/releases/download/v1.2/GhostFaceNet_W1.3_S1_ArcFace.h5"
class GhostFaceNetClient(FacialRecognition):
@ -71,12 +71,10 @@ def load_model():
model = GhostFaceNetV1()
weight_file = weight_utils.download_weights_if_necessary(
file_name="ghostfacenet_v1.h5", source_url=PRETRAINED_WEIGHTS
file_name="ghostfacenet_v1.h5", source_url=WEIGHTS_URL
)
model = weight_utils.load_model_weights(
model=model, weight_file=weight_file
)
model = weight_utils.load_model_weights(model=model, weight_file=weight_file)
return model

4
deepface/models/facial_recognition/OpenFace.py
View File

@ -24,6 +24,8 @@ else:
# pylint: disable=unnecessary-lambda
WEIGHTS_URL="https://github.com/serengil/deepface_models/releases/download/v1.0/openface_weights.h5"
# ---------------------------------------
# pylint: disable=too-few-public-methods
@ -40,7 +42,7 @@ class OpenFaceClient(FacialRecognition):
def load_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/openface_weights.h5",
url=WEIGHTS_URL,
) -> Model:
"""
Consturct OpenFace model, download its weights and load

3
deepface/models/facial_recognition/SFace.py
View File

@ -13,6 +13,7 @@ from deepface.commons.logger import Logger
logger = Logger()
# pylint: disable=line-too-long, too-few-public-methods
WEIGHTS_URL = "https://github.com/opencv/opencv_zoo/raw/main/models/face_recognition_sface/face_recognition_sface_2021dec.onnx"
class SFaceClient(FacialRecognition):
@ -47,7 +48,7 @@ class SFaceClient(FacialRecognition):
def load_model(
url="https://github.com/opencv/opencv_zoo/raw/main/models/face_recognition_sface/face_recognition_sface_2021dec.onnx",
url=WEIGHTS_URL,
) -> Any:
"""
Construct SFace model, download its weights and load

11
deepface/models/facial_recognition/VGGFace.py
View File

@ -38,6 +38,10 @@ else:
# ---------------------------------------
WEIGHTS_URL = (
"https://github.com/serengil/deepface_models/releases/download/v1.0/vgg_face_weights.h5"
)
# pylint: disable=too-few-public-methods
class VggFaceClient(FacialRecognition):
"""
@ -126,7 +130,7 @@ def base_model() -> Sequential:
def load_model(
url="https://github.com/serengil/deepface_models/releases/download/v1.0/vgg_face_weights.h5",
url=WEIGHTS_URL,
) -> Model:
"""
Final VGG-Face model being used for finding embeddings
@ -140,9 +144,7 @@ def load_model(
file_name="vgg_face_weights.h5", source_url=url
)
model = weight_utils.load_model_weights(
model=model, weight_file=weight_file
)
model = weight_utils.load_model_weights(model=model, weight_file=weight_file)
# 2622d dimensional model
# vgg_face_descriptor = Model(inputs=model.layers[0].input, outputs=model.layers[-2].output)
@ -151,7 +153,6 @@ def load_model(
# - softmax causes underfitting
# - added normalization layer to avoid underfitting with euclidean
# as described here: https://github.com/serengil/deepface/issues/944
base_model_output = Sequential()
base_model_output = Flatten()(model.layers[-5].output)
# keras backend's l2 normalization layer troubles some gpu users (e.g. issue 957, 966)
# base_model_output = Lambda(lambda x: K.l2_normalize(x, axis=1), name="norm_layer")(

7
deepface/models/spoofing/FasNet.py
View File

@ -12,6 +12,9 @@ from deepface.commons.logger import Logger
logger = Logger()
# pylint: disable=line-too-long, too-few-public-methods, nested-min-max
FIRST_WEIGHTS_URL="https://github.com/minivision-ai/Silent-Face-Anti-Spoofing/raw/master/resources/anti_spoof_models/2.7_80x80_MiniFASNetV2.pth"
SECOND_WEIGHTS_URL="https://github.com/minivision-ai/Silent-Face-Anti-Spoofing/raw/master/resources/anti_spoof_models/4_0_0_80x80_MiniFASNetV1SE.pth"
class Fasnet:
"""
Mini Face Anti Spoofing Net Library from repo: github.com/minivision-ai/Silent-Face-Anti-Spoofing
@ -35,12 +38,12 @@ class Fasnet:
# download pre-trained models if not installed yet
first_model_weight_file = weight_utils.download_weights_if_necessary(
file_name="2.7_80x80_MiniFASNetV2.pth",
source_url="https://github.com/minivision-ai/Silent-Face-Anti-Spoofing/raw/master/resources/anti_spoof_models/2.7_80x80_MiniFASNetV2.pth",
source_url=FIRST_WEIGHTS_URL,
)
second_model_weight_file = weight_utils.download_weights_if_necessary(
file_name="4_0_0_80x80_MiniFASNetV1SE.pth",
source_url="https://github.com/minivision-ai/Silent-Face-Anti-Spoofing/raw/master/resources/anti_spoof_models/4_0_0_80x80_MiniFASNetV1SE.pth",
source_url=SECOND_WEIGHTS_URL,
)
# guarantees Fasnet imported and torch installed

46
deepface/modules/detection.py
View File

@ -148,16 +148,26 @@ def extract_faces(
w = min(width - x - 1, int(current_region.w))
h = min(height - y - 1, int(current_region.h))
facial_area = {
"x": x,
"y": y,
"w": w,
"h": h,
"left_eye": current_region.left_eye,
"right_eye": current_region.right_eye,
}
# optional nose, mouth_left and mouth_right fields are coming just for retinaface
if current_region.nose is not None:
facial_area["nose"] = current_region.nose
if current_region.mouth_left is not None:
facial_area["mouth_left"] = current_region.mouth_left
if current_region.mouth_right is not None:
facial_area["mouth_right"] = current_region.mouth_right
resp_obj = {
"face": current_img,
"facial_area": {
"x": x,
"y": y,
"w": w,
"h": h,
"left_eye": current_region.left_eye,
"right_eye": current_region.right_eye,
},
"facial_area": facial_area,
"confidence": round(float(current_region.confidence or 0), 2),
}
@ -272,6 +282,9 @@ def expand_and_align_face(
left_eye = facial_area.left_eye
right_eye = facial_area.right_eye
confidence = facial_area.confidence
nose = facial_area.nose
mouth_left = facial_area.mouth_left
mouth_right = facial_area.mouth_right
if expand_percentage > 0:
# Expand the facial region height and width by the provided percentage
@ -305,11 +318,26 @@ def expand_and_align_face(
left_eye = (left_eye[0] - width_border, left_eye[1] - height_border)
if right_eye is not None:
right_eye = (right_eye[0] - width_border, right_eye[1] - height_border)
if nose is not None:
nose = (nose[0] - width_border, nose[1] - height_border)
if mouth_left is not None:
mouth_left = (mouth_left[0] - width_border, mouth_left[1] - height_border)
if mouth_right is not None:
mouth_right = (mouth_right[0] - width_border, mouth_right[1] - height_border)
return DetectedFace(
img=detected_face,
facial_area=FacialAreaRegion(
x=x, y=y, h=h, w=w, confidence=confidence, left_eye=left_eye, right_eye=right_eye
x=x,
y=y,
h=h,
w=w,
confidence=confidence,
left_eye=left_eye,
right_eye=right_eye,
nose=nose,
mouth_left=mouth_left,
mouth_right=mouth_right,
),
confidence=confidence,
)

115
deepface/modules/recognition.py
View File

@ -266,7 +266,7 @@ def find(
align,
threshold,
normalization,
anti_spoofing
anti_spoofing,
)
df = pd.DataFrame(representations)
@ -441,6 +441,7 @@ def __find_bulk_embeddings(
return representations
def find_batched(
representations: List[Dict[str, Any]],
source_objs: List[Dict[str, Any]],
@ -508,27 +509,24 @@ def find_batched(
metadata = set()
for item in representations:
emb = item.get('embedding')
emb = item.get("embedding")
if emb is not None:
embeddings_list.append(emb)
valid_mask.append(True)
else:
embeddings_list.append(np.zeros_like(representations[0]['embedding']))
embeddings_list.append(np.zeros_like(representations[0]["embedding"]))
valid_mask.append(False)
metadata.update(item.keys())
# remove embedding key from other keys
metadata.discard('embedding')
metadata.discard("embedding")
metadata = list(metadata)
embeddings = np.array(embeddings_list) # (N, D)
valid_mask = np.array(valid_mask) # (N,)
embeddings = np.array(embeddings_list) # (N, D)
valid_mask = np.array(valid_mask) # (N,)
data = {
key: np.array([item.get(key, None) for item in representations])
for key in metadata
}
data = {key: np.array([item.get(key, None) for item in representations]) for key in metadata}
target_embeddings = []
source_regions = []
@ -558,101 +556,46 @@ def find_batched(
target_threshold = threshold or verification.find_threshold(model_name, distance_metric)
target_thresholds.append(target_threshold)
target_embeddings = np.array(target_embeddings) # (M, D)
target_thresholds = np.array(target_thresholds) # (M,)
target_embeddings = np.array(target_embeddings) # (M, D)
target_thresholds = np.array(target_thresholds) # (M,)
source_regions_arr = {
'source_x': np.array([region['x'] for region in source_regions]),
'source_y': np.array([region['y'] for region in source_regions]),
'source_w': np.array([region['w'] for region in source_regions]),
'source_h': np.array([region['h'] for region in source_regions]),
"source_x": np.array([region["x"] for region in source_regions]),
"source_y": np.array([region["y"] for region in source_regions]),
"source_w": np.array([region["w"] for region in source_regions]),
"source_h": np.array([region["h"] for region in source_regions]),
}
def find_cosine_distance_batch(
embeddings: np.ndarray, target_embeddings: np.ndarray
) -> np.ndarray:
"""
Find the cosine distances between batches of embeddings
Args:
embeddings (np.ndarray): array of shape (N, D)
target_embeddings (np.ndarray): array of shape (M, D)
Returns:
np.ndarray: distance matrix of shape (M, N)
"""
embeddings_norm = verification.l2_normalize(embeddings, axis=1)
target_embeddings_norm = verification.l2_normalize(target_embeddings, axis=1)
cosine_similarities = np.dot(target_embeddings_norm, embeddings_norm.T)
cosine_distances = 1 - cosine_similarities
return cosine_distances
def find_euclidean_distance_batch(
embeddings: np.ndarray, target_embeddings: np.ndarray
) -> np.ndarray:
"""
Find the Euclidean distances between batches of embeddings
Args:
embeddings (np.ndarray): array of shape (N, D)
target_embeddings (np.ndarray): array of shape (M, D)
Returns:
np.ndarray: distance matrix of shape (M, N)
"""
diff = embeddings[None, :, :] - target_embeddings[:, None, :] # (M, N, D)
distances = np.linalg.norm(diff, axis=2) # (M, N)
return distances
def find_distance_batch(
embeddings: np.ndarray, target_embeddings: np.ndarray, distance_metric: str,
) -> np.ndarray:
"""
Find pairwise distances between batches of embeddings using the specified distance metric
Args:
embeddings (np.ndarray): array of shape (N, D)
target_embeddings (np.ndarray): array of shape (M, D)
distance_metric (str): distance metric ('cosine', 'euclidean', 'euclidean_l2')
Returns:
np.ndarray: distance matrix of shape (M, N)
"""
if distance_metric == "cosine":
distances = find_cosine_distance_batch(embeddings, target_embeddings)
elif distance_metric == "euclidean":
distances = find_euclidean_distance_batch(embeddings, target_embeddings)
elif distance_metric == "euclidean_l2":
embeddings_norm = verification.l2_normalize(embeddings, axis=1)
target_embeddings_norm = verification.l2_normalize(target_embeddings, axis=1)
distances = find_euclidean_distance_batch(embeddings_norm, target_embeddings_norm)
else:
raise ValueError("Invalid distance_metric passed - ", distance_metric)
return np.round(distances, 6)
distances = find_distance_batch(embeddings, target_embeddings, distance_metric) # (M, N)
distances = verification.find_distance(embeddings, target_embeddings, distance_metric) # (M, N)
distances[:, ~valid_mask] = np.inf
resp_obj = []
for i in range(len(target_embeddings)):
target_distances = distances[i] # (N,)
target_distances = distances[i] # (N,)
target_threshold = target_thresholds[i]
N = embeddings.shape[0]
result_data = dict(data)
result_data.update({
'source_x': np.full(N, source_regions_arr['source_x'][i]),
'source_y': np.full(N, source_regions_arr['source_y'][i]),
'source_w': np.full(N, source_regions_arr['source_w'][i]),
'source_h': np.full(N, source_regions_arr['source_h'][i]),
'threshold': np.full(N, target_threshold),
'distance': target_distances,
})
result_data.update(
{
"source_x": np.full(N, source_regions_arr["source_x"][i]),
"source_y": np.full(N, source_regions_arr["source_y"][i]),
"source_w": np.full(N, source_regions_arr["source_w"][i]),
"source_h": np.full(N, source_regions_arr["source_h"][i]),
"threshold": np.full(N, target_threshold),
"distance": target_distances,
}
)
mask = target_distances <= target_threshold
filtered_data = {key: value[mask] for key, value in result_data.items()}
sorted_indices = np.argsort(filtered_data['distance'])
sorted_indices = np.argsort(filtered_data["distance"])
sorted_data = {key: value[sorted_indices] for key, value in filtered_data.items()}
num_results = len(sorted_data['distance'])
num_results = len(sorted_data["distance"])
result_dicts = [
{key: sorted_data[key][i] for key in sorted_data}
for i in range(num_results)
{key: sorted_data[key][i] for key in sorted_data} for i in range(num_results)
]
resp_obj.append(result_dicts)
return resp_obj

118
deepface/modules/verification.py
View File

@ -263,45 +263,73 @@ def __extract_faces_and_embeddings(
def find_cosine_distance(
source_representation: Union[np.ndarray, list], test_representation: Union[np.ndarray, list]
) -> np.float64:
) -> Union[np.float64, np.ndarray]:
"""
Find cosine distance between two given vectors
Find cosine distance between two given vectors or batches of vectors.
Args:
source_representation (np.ndarray or list): 1st vector
test_representation (np.ndarray or list): 2nd vector
source_representation (np.ndarray or list): 1st vector or batch of vectors.
test_representation (np.ndarray or list): 2nd vector or batch of vectors.
Returns
distance (np.float64): calculated cosine distance
np.float64 or np.ndarray: Calculated cosine distance(s).
It returns a np.float64 for single embeddings and np.ndarray for batch embeddings.
"""
if isinstance(source_representation, list):
source_representation = np.array(source_representation)
# Convert inputs to numpy arrays if necessary
source_representation = np.asarray(source_representation)
test_representation = np.asarray(test_representation)
if isinstance(test_representation, list):
test_representation = np.array(test_representation)
a = np.dot(source_representation, test_representation)
b = np.linalg.norm(source_representation)
c = np.linalg.norm(test_representation)
return 1 - a / (b * c)
if source_representation.ndim == 1 and test_representation.ndim == 1:
# single embedding
dot_product = np.dot(source_representation, test_representation)
source_norm = np.linalg.norm(source_representation)
test_norm = np.linalg.norm(test_representation)
distances = 1 - dot_product / (source_norm * test_norm)
elif source_representation.ndim == 2 and test_representation.ndim == 2:
# list of embeddings (batch)
source_normed = l2_normalize(source_representation, axis=1) # (N, D)
test_normed = l2_normalize(test_representation, axis=1) # (M, D)
cosine_similarities = np.dot(test_normed, source_normed.T) # (M, N)
distances = 1 - cosine_similarities
else:
raise ValueError(
f"Embeddings must be 1D or 2D, but received "
f"source shape: {source_representation.shape}, test shape: {test_representation.shape}"
)
return distances
def find_euclidean_distance(
source_representation: Union[np.ndarray, list], test_representation: Union[np.ndarray, list]
) -> np.float64:
) -> Union[np.float64, np.ndarray]:
"""
Find euclidean distance between two given vectors
Find Euclidean distance between two vectors or batches of vectors.
Args:
source_representation (np.ndarray or list): 1st vector
test_representation (np.ndarray or list): 2nd vector
Returns
distance (np.float64): calculated euclidean distance
source_representation (np.ndarray or list): 1st vector or batch of vectors.
test_representation (np.ndarray or list): 2nd vector or batch of vectors.
Returns:
np.float64 or np.ndarray: Euclidean distance(s).
Returns a np.float64 for single embeddings and np.ndarray for batch embeddings.
"""
if isinstance(source_representation, list):
source_representation = np.array(source_representation)
# Convert inputs to numpy arrays if necessary
source_representation = np.asarray(source_representation)
test_representation = np.asarray(test_representation)
if isinstance(test_representation, list):
test_representation = np.array(test_representation)
return np.linalg.norm(source_representation - test_representation)
# Single embedding case (1D arrays)
if source_representation.ndim == 1 and test_representation.ndim == 1:
distances = np.linalg.norm(source_representation - test_representation)
# Batch embeddings case (2D arrays)
elif source_representation.ndim == 2 and test_representation.ndim == 2:
diff = (
source_representation[None, :, :] - test_representation[:, None, :]
) # (N, D) - (M, D) = (M, N, D)
distances = np.linalg.norm(diff, axis=2) # (M, N)
else:
raise ValueError(
f"Embeddings must be 1D or 2D, but received "
f"source shape: {source_representation.shape}, test shape: {test_representation.shape}"
)
return distances
def l2_normalize(
@ -315,8 +343,8 @@ def l2_normalize(
Returns:
np.ndarray: l2 normalized vector
"""
if isinstance(x, list):
x = np.array(x)
# Convert inputs to numpy arrays if necessary
x = np.asarray(x)
norm = np.linalg.norm(x, axis=axis, keepdims=True)
return x / (norm + epsilon)
@ -325,23 +353,39 @@ def find_distance(
alpha_embedding: Union[np.ndarray, list],
beta_embedding: Union[np.ndarray, list],
distance_metric: str,
) -> np.float64:
) -> Union[np.float64, np.ndarray]:
"""
Wrapper to find distance between vectors according to the given distance metric
Wrapper to find the distance between vectors based on the specified distance metric.
Args:
source_representation (np.ndarray or list): 1st vector
test_representation (np.ndarray or list): 2nd vector
Returns
distance (np.float64): calculated cosine distance
alpha_embedding (np.ndarray or list): 1st vector or batch of vectors.
beta_embedding (np.ndarray or list): 2nd vector or batch of vectors.
distance_metric (str): The type of distance to compute
('cosine', 'euclidean', or 'euclidean_l2').
Returns:
np.float64 or np.ndarray: The calculated distance(s).
"""
# Convert inputs to numpy arrays if necessary
alpha_embedding = np.asarray(alpha_embedding)
beta_embedding = np.asarray(beta_embedding)
# Ensure that both embeddings are either 1D or 2D
if alpha_embedding.ndim != beta_embedding.ndim or alpha_embedding.ndim not in (1, 2):
raise ValueError(
f"Both embeddings must be either 1D or 2D, but received "
f"alpha shape: {alpha_embedding.shape}, beta shape: {beta_embedding.shape}"
)
if distance_metric == "cosine":
distance = find_cosine_distance(alpha_embedding, beta_embedding)
elif distance_metric == "euclidean":
distance = find_euclidean_distance(alpha_embedding, beta_embedding)
elif distance_metric == "euclidean_l2":
distance = find_euclidean_distance(
l2_normalize(alpha_embedding), l2_normalize(beta_embedding)
)
axis = None if alpha_embedding.ndim == 1 else 1
normalized_alpha = l2_normalize(alpha_embedding, axis=axis)
normalized_beta = l2_normalize(beta_embedding, axis=axis)
distance = find_euclidean_distance(normalized_alpha, normalized_beta)
else:
raise ValueError("Invalid distance_metric passed - ", distance_metric)
return np.round(distance, 6)