Recognize and manipulate faces from Python or from the command line with
the world's simplest face recognition library.
Built using dlib's state-of-the-art face recognition
built with deep learning. The model has an accuracy of 99.38% on the
Labeled Faces in the Wild benchmark (see the LFW paper (PDF)).
This also provides a simple face_recognition command line tool that lets
you do face recognition on a folder of images from the command line!
Finding facial features is super useful for lots of important stuff. But you can also use it for really stupid stuff
like applying digital make-up (think 'Meitu'):
Please follow the instructions in the article carefully. There is current a bug in the CUDA libraries on the Jetson Nano that will cause this library to fail silently if you don't follow the instructions in the article to comment out a line in dlib and recompile it.
When you install face_recognition, you get two simple command-line
programs:
face_recognition - Recognize faces in a photograph or folder full for
photographs.
face_detection - Find faces in a photograph or folder full for photographs.
face_recognition command line tool
The face_recognition command lets you recognize faces in a photograph or
folder full for photographs.
First, you need to provide a folder with one picture of each person you
already know. There should be one image file for each person with the
files named according to who is in the picture:
Next, you need a second folder with the files you want to identify:
Then in you simply run the command face_recognition, passing in
the folder of known people and the folder (or single image) with unknown
people and it tells you who is in each image:
It prints one line for each face that was detected. The coordinates
reported are the top, right, bottom and left coordinates of the face (in pixels).
Adjusting Tolerance / Sensitivity
If you are getting multiple matches for the same person, it might be that
the people in your photos look very similar and a lower tolerance value
is needed to make face comparisons more strict.
You can do that with the --tolerance parameter. The default tolerance
value is 0.6 and lower numbers make face comparisons more strict:
Face recognition can be done in parallel if you have a computer with
multiple CPU cores. For example, if your system has 4 CPU cores, you can
process about 4 times as many images in the same amount of time by using
all your CPU cores in parallel.
If you are using Python 3.4 or newer, pass in a --cpus <number_of_cpu_cores_to_use> parameter:
import face_recognition
image = face_recognition.load_image_file("my_picture.jpg")
face_locations = face_recognition.face_locations(image)
# face_locations is now an array listing the co-ordinates of each face!
You can also opt-in to a somewhat more accurate deep-learning-based face detection model.
Note: GPU acceleration (via NVidia's CUDA library) is required for good
performance with this model. You'll also want to enable CUDA support
when compliling dlib.
import face_recognition
image = face_recognition.load_image_file("my_picture.jpg")
face_locations = face_recognition.face_locations(image, model="cnn")
# face_locations is now an array listing the co-ordinates of each face!
Automatically locate the facial features of a person in an image
import face_recognition
image = face_recognition.load_image_file("my_picture.jpg")
face_landmarks_list = face_recognition.face_landmarks(image)
# face_landmarks_list is now an array with the locations of each facial feature in each face.
# face_landmarks_list[0]['left_eye'] would be the location and outline of the first person's left eye.
Recognize faces in images and identify who they are
import face_recognition
picture_of_me = face_recognition.load_image_file("me.jpg")
my_face_encoding = face_recognition.face_encodings(picture_of_me)[0]
# my_face_encoding now contains a universal 'encoding' of my facial features that can be compared to any other picture of a face!
unknown_picture = face_recognition.load_image_file("unknown.jpg")
unknown_face_encoding = face_recognition.face_encodings(unknown_picture)[0]
# Now we can see the two face encodings are of the same person with `compare_faces`!
results = face_recognition.compare_faces([my_face_encoding], unknown_face_encoding)
if results[0] == True:
print("It's a picture of me!")
else:
print("It's not a picture of me!")
If you want to create a standalone executable that can run without the need to install python or face_recognition, you can use PyInstaller. However, it requires some custom configuration to work with this library. See this issue for how to do it.
Covers how to automatically cluster photos based on who appears in each photo using unsupervised learning
How Face Recognition Works
If you want to learn how face location and recognition work instead of
depending on a black box library, read my article.
Caveats
The face recognition model is trained on adults and does not work very well on children. It tends to mix
up children quite easy using the default comparison threshold of 0.6.
Accuracy may vary between ethnic groups. Please see this wiki page for more details.
<a name="deployment">Deployment to Cloud Hosts (Heroku, AWS, etc)</a>
Since face_recognition depends on dlib which is written in C++, it can be tricky to deploy an app
using it to a cloud hosting provider like Heroku or AWS.
To make things easier, there's an example Dockerfile in this repo that shows how to run an app built with
face_recognition in a Docker container. With that, you should be able to deploy
to any service that supports Docker images.
You can try the Docker image locally by running: docker-compose up --build
Linux users with a GPU (drivers >= 384.81) and Nvidia-Docker installed can run the example on the GPU: Open the docker-compose.yml file and uncomment the dockerfile: Dockerfile.gpu and runtime: nvidia lines.
Having problems?
If you run into problems, please read the Common Errors section of the wiki before filing a github issue.
Thanks
Many, many thanks to Davis King (@nulhom)
for creating dlib and for providing the trained facial feature detection and face encoding models
used in this library. For more information on the ResNet that powers the face encodings, check out
his blog post.
Thanks to everyone who works on all the awesome Python data science libraries like numpy, scipy, scikit-image,
pillow, etc, etc that makes this kind of stuff so easy and fun in Python.
