Security News
tea.xyz Spam Plagues npm and RubyGems Package Registries
Tea.xyz, a crypto project aimed at rewarding open source contributions, is once again facing backlash due to an influx of spam packages flooding public package registries.
Readme
The fdet
is a ready-to-use implementation of deep learning face detectors using PyTorch.
You can use it directly in your code, as a python library:
>>> from fdet import io, RetinaFace
>>> detector = RetinaFace(backbone='RESNET50')
>>> detector.detect(io.read_as_rgb('path_to_image.jpg'))
[{'box': [511, 47, 35, 45], 'confidence': 0.9999996423721313,
'keypoints': {'left_eye': [517, 70], 'right_eye': [530, 65], 'nose': [520, 77],
'mouth_left': [522, 87], 'mouth_right': [531, 83]}}]
Or through command-line application:
fdet retinaface -b RESNET50 -i path_to_image.jpg -o detections.json
Currently, there are two different detectors available on FDet:
Despite the availability of different implementations of these algorithms, there are some disadvantages we found when using them. So we create this project to offer the following features, in one package:
You need to install PyTorch first (if you have a GPU, install PyTorch with CUDA support).
Then fdet
can be installed through pip:
pip install fdet
If you want to use fdet
from python, just import it,
from fdet import MTCNN, RetinaFace
and instantiate your desired detector, with its respective parameters:
MTCNN(thresholds, nms_thresholds, min_face_size, cuda_enable, cuda_devices)
thresholds
(tuple, optional): The thresholds fo each MTCNN step [default: (0.6, 0.7, 0.8)]nms_thresholds
(tuple, optional): The NMS thresholds fo each MTCNN step [default: (0.7, 0.7, 0.7)]min_face_size
(float, optional): The minimum size of the face to detect, in pixels [default: 20.0].cuda_enable
(bool, optional): Indicates wheter CUDA, if available, should be used or not. If False, uses only CPU processing [default: True].cuda_devices
(list, optional): List of CUDA GPUs to be used. If None, uses all avaliable GPUs [default: None]. If cuda_enable
is False, this parameter is ignored.RetinaFace(backbone, threshold, nms_threshold, max_face_size, cuda_enable, cuda_devices)
backbone
(str): The backbone model ['RESNET50'
or 'MOBILENET'
].threshold
(tuple, optional): The detection threshold [default: 0.8]nms_threshold
(tuple, optional): The NMS threshold [default: 0.4]max_face_size
(int, optional): The maximum size of the face to detect, in pixels [default: 1000].cuda_enable
(bool, optional): Indicates wheter CUDA, if available, should be used or not. If False, uses only CPU processing. [default: True].cuda_devices
(list, optional): List of CUDA GPUs to be used. If None, uses all avaliable GPUs. [default: None]. If cuda_enable
is False, this parameter is ignored.The detector classes listed above have two methods to perform the face detection:
The following example illustrates the ease of use of this library to detect faces in a single-image using MTCNN and to create an output image with all detected faces drawn.
>>> from fdet import io, MTCNN
>>>
>>> detector = MTCNN()
>>> image = io.read_as_rgb('example.jpg')
>>>
>>> detections = detector.detect(image)
>>> print(detections)
[{'box': [511, 47, 35, 45], 'confidence': 0.9999996423721313,
'keypoints': {'left_eye': [517, 70], 'right_eye': [530, 65], 'nose': [520, 77],
'mouth_left': [522, 87], 'mouth_right': [531, 83]}}]
>>>
>>> output_image = io.draw_detections(image, detections, color='white', thickness=5)
>>> io.save('output.jpg', output_image)
The detector returns a list of dict
objects. Each dict
contains three main keys:
'box'
: The bounding box formatted as a list [x, y, width, height]
;'confidence'
: The probability for a bounding box to be matching a face;'keypoints'
:
The five landmarks formatted into a dict
with the keys 'left_eye'
, 'right_eye'
, 'nose'
, 'mouth_left'
, 'mouth_right'
. Each keypoint is identified by a pixel position [x, y]
.The io.read_as_rgb()
is a wrapper for opencv cv2.imread()
to ensure an RGB image and can be replaced by:
image = cv2.imread('example.jpg', cv2.IMREAD_COLOR)
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
The library is also capable of performing face detection on image batches, typically providing considerable speed-up. A batch should be structured as list of images (numpy
arrays) of equal dimension. The returned detections list will have an additional first dimension corresponding to the batch size. Each image in the batch may have one or more faces detected.
In the following example, we detect faces in every frame of a video using batchs of 10 images.
>>> import cv2
>>> from fdet import io, RetinaFace
>>>
>>> BATCH_SIZE = 10
>>>
>>> detector = RetinaFace(backbone='MOBILENET', cuda_devices=[0,1])
>>> vid_cap = cv2.VideoCapture('path_to_video.mp4')
>>>
>>> video_face_detections = [] # list to store all video face detections
>>> image_buffer = [] # buffer to store the batch
>>>
>>> while True:
>>>
>>> success, frame = vid_cap.read() # read the frame from video capture
>>> if not success:
>>> break # end of video
>>>
>>> frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # convert to RGB
>>> image_buffer.append(frame) # add frame to buffer
>>>
>>> if len(image_buffer) == BATCH_SIZE: # if buffer is full, detect the batch
>>> batch_detections = detector.batch_detect(image_buffer)
>>> video_face_detections.extend(batch_detections)
>>> image_buffer.clear() # clear the buffer
>>>
>>> if image_buffer: # checks if images remain in the buffer and detect it
>>> batch_detections = detector.batch_detect(image_buffer)
>>> video_face_detections.extend(batch_detections)
The FDet was written heavily inspired by the other available implementations (see credits).
The current MTCNN version was implemented with the help of Davi Beltrão.
[zhang:2016]: Zhang, K., Zhang, Z., Li, Z. and Qiao, Y. (2016). Joint face detection and alignment using multitask cascaded convolutional networks. IEEE Signal Processing Letters, 23(10), 1499-1503. (link to paper)
[deng:2019]: Deng, J., Guo, J., Zhou, Y., Yu, J., Kotsia, I. and Zafeiriou, S. (2019). Retinaface: Single-stage dense face localisation in the wild. arXiv preprint arXiv:1905.00641. (link to paper)
FAQs
An easy to use face detection module based on MTCNN and RetinaFace algorithms.
We found that fdet demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Security News
Tea.xyz, a crypto project aimed at rewarding open source contributions, is once again facing backlash due to an influx of spam packages flooding public package registries.
Security News
As cyber threats become more autonomous, AI-powered defenses are crucial for businesses to stay ahead of attackers who can exploit software vulnerabilities at scale.
Security News
UnitedHealth Group disclosed that the ransomware attack on Change Healthcare compromised protected health information for millions in the U.S., with estimated costs to the company expected to reach $1 billion.