Manuel Rios & Tomas Correa
This folder contains the development of final project for the Computer Vision course at Universidad de los Andes.
Experiments are configured to run smoothly on BCV002 machine. Dataset is available in the following path:
media/disk0/Datasets_FP/Correa_Rios. Throughout a symbolic link, all permissions are granted.
All execution modes will use our best method FaceNet with triplet and classification loss.
All default arguments can be found inside main.py.
To train run the following command:
python main.py --mode train
Optional arguments are learning rate, number of epochs, and batch size:
python main.py --mode train --lr 0.1 --num_epochs 16 --bs 48
Testing should be done with the following command:
python main.py --mode test
Demo should be done with the following command:
python main.py --mode demo --pair_type md
Where the pair_type argument indicates the kinship relation to validate (choices are fs, fd, ms, md).
Optional arguments are the test images to use, requiring only to specify the test image file name (not whole path):
python main.py --mode demo --pair_type md --img1 face1.jpg --img2 face2.jpg
For the demo, a positive verification will print "KIN: pair_type" in the terminal. A negative result prints "NO KIN". Both results print the predicted probability of kinship.