Make sure you have python, OpenFace and dlib installed. You can either install them manually or use a preconfigured docker image that has everying already installed:
docker pull bamos/openface
docker run -p 9000:9000 -p 8000:8000 -t -i bamos/openface /bin/bash
cd /root/openfacePro-tip: If you are using Docker on OSX, you can make your OSX /Users/ folder visible inside a docker image like this:
docker run -v /Users:/host/Users -p 9000:9000 -p 8000:8000 -t -i bamos/openface /bin/bash
cd /root/openface
Then you can access all your OSX files inside of the docker image at /host/Users/...
ls /host/Users/
Make a folder called ./training-images/ inside the openface folder.
mkdir training-images
Make a subfolder for each person you want to recognize. For example:
mkdir ./training-images/will-ferrell/
mkdir ./training-images/chad-smith/
mkdir ./training-images/jimmy-fallon/
Copy all your images of each person into the correct sub-folders. Make sure only one face appears in each image. There's no need to crop the image around the face. OpenFace will do that automatically.
Run the openface scripts from inside the openface root directory:
First, do pose detection and alignment:
./util/align-dlib.py ./training-images/ align outerEyesAndNose ./aligned-images/ --size 96
This will create a new ./aligned-images/ subfolder with a cropped and aligned version of each of your test images.
Second, generate the representations from the aligned images:
./batch-represent/main.lua -outDir ./generated-embeddings/ -data ./aligned-images/
After you run this, the ./generated-embeddings/ sub-folder will contain a csv file with the embeddings for each image.
Third, train your face detection model:
./demos/classifier.py train ./generated-embeddings/
This will generate a new file called ./generated-embeddings/classifier.pkl. This file has the SVM model you'll use to recognize
new faces.
At this point, you should have a working face recognizer!
Get a new picture with an unknown face. Pass it to the classifier script like this:
./demos/classifier.py infer ./generated-embeddings/classifier.pkl your_test_image.jpg
You should get a prediction that looks like this:
=== /test-images/will-ferrel-1.jpg ===
Predict will-ferrell with 0.73 confidence.
From here it's up to you to adapt the ./demos/classifier.py python script to work however you want.
Important notes:
- If you get bad results, try adding a few more pictures of each person in Step 3 (especially picures in different poses).
- This script will always make a prediction even if the face isn't one it knows. In a real application, you would look at the confidence score and throw away predictions with a low confidence since they are most likely wrong.
Can I use this on raspberry pi ?