This is a repo forked from http://travis-ci.org/davidsandberg/facenet.
I have modified the python commands for the Tensorflow implementation to be compatible with Windows. You can see the changes by comparing my repo with the original repo.
Tensorflow version used: 1.13.0
Libraries Required: scipy matplotlib scikit-learn
| Version Date | Update |
|---|---|
| 2018-04-10 | Added new models trained on Casia-WebFace and VGGFace2 (see below). Note that the models uses fixed image standardization (see [wiki] |
I have used the pre-trained model below in my code - you will need to create your own folder 'models' and change the path accordingly in data/test.py.
| Model name | LFW accuracy | Training dataset | Architecture |
|---|---|---|---|
| 20180402-114759 | 0.9965 | VGGFace2 | Inception ResNet v1 |
To align the datasets using MTCNN:
python C:/Users/sonur/Downloads/Bio/facenet-master/src/align/align_dataset_mtcnn.py "C:/Users/sonur/Downloads/Bio/Dataset/Chang/Blur" "C:/Users/sonur/Downloads/Bio/Dataset/Changprocessed/BlurProcessed" --image_size 160 --margin 32 --random_order --gpu_memory_fraction 0.25
where
"C:/Users/sonur/Downloads/Bio/Dataset/lfw/raw – path to images
C:/Users/sonur/Downloads/Bio/Dataset/processed – path where you want aligned images to be stored
To evaluate the results on the LFW dataset:
python C:/Users/sonur/Downloads/Bio/facenet-master/src/validate_on_lfw.py "C:/Users/sonur/Downloads/Bio/Dataset/Changprocessed/BlurProcessed" "C:/Users/sonur/Downloads/Bio/facenet-master/models/20180402-114759" --distance_metric 1 --use_flipped_images --subtract_mean --use_fixed_image_standardization
where
"C:/Users/sonur/Downloads/Bio/facenet-master/models/20180402-114759" is the path to the pre-trained model
To run the test use the description found on the page Validate on LFW. Note that the input images to the model need to be standardized using fixed image standardization (use the option --use_fixed_image_standardization when running e.g. validate_on_lfw.py).