Facial Recognition with Tensorflow

Tensorflow pipeline to recognize faces for python 3.6.

Slides

Google Slides

1. Installing Dependencies

FaceNet

Run the following command.

git clone https://github.com/davidsandberg/facenet.git
export PYTHONPATH=~/<facenet_path>/src:~/<facenet_path>/contributed

• Replace <facenet_path> with the facenet installation folder.

Python modules

Run the following command.

pip install -r requirements.txt

2. Download resources

• LFW dataset

• Model Checkpoint

3. Align the LFW dataset

Run the following command.

for N in {1..4}; do python ~/<facenet_path>/src/align/align_dataset_mtcnn.py ~/<lfw_path>/raw ~/<lfw_path>/lfw_mtcnnpy_160 --image_size 160 --margin 32 --random_order --gpu_memory_fraction 0.25 & done

• Replace <facenet_path> with the facenet installation folder.
• Replace <lfw_path> with the lfw installation folder.

4. Copy the aligned faces to the dataset

Run the following command.

cp ~/<lfw_path>/lfw_mtcnnpy_160/* ~/<repo_path>/datasets

• Replace <repo_path> with this repository installation folder.
• Replace <lfw_path> with the lfw installation folder.

5. Add a new person

Run the following command.

~/<repo_path>/bin/add_new_face

• Replace <repo_path> with this repository installation folder.

6. Train a classifier

Run the following command.

python ~/<facenet_path>/src/classifier.py TRAIN ~/<repo_path>/datasets ~/<repo_path>/models/20170512-110547/20170512-110547.pb ~/<repo_path>/classifier/face_classifier.pkl --batch_size 1000 --min_nrof_images_per_class 40 --nrof_train_images_per_class 40

• Replace <facenet_path> with the facenet installation folder.
• Replace <repo_path> with this repository installation folder.

7. Start the webserver

Run the following command.

python server.py

Credits

• FaceNet Repository
• Multi-task CNN
• "FaceNet: A Unified Embedding for Face Recognition and Clustering"

License

MIT License