Face recognition using few shot learning

This is the PARAN semester project in Ajou university in 2021 spring.

The double relation few-shot model can recognition someone's face faster than other deep learning model and the model size is very small. You can use this model only 3 step.

1. Register your face in facebank

We call face database a facebank and it has the following folder structure.

facebank
`-- images
    |-- hankyul
    |   |-- 1.jpg
    |   `-- 2.jpg
    |-- sanghyun
    |   |-- 1.jpg
    |   `-- 2.jpg
    `-- seungmin
        |-- 1.jpg
        `-- 2.jpg

In facebank, you can register someone's class and face.

2. Make facebank vector

For fast face recognition, you have to make facebank vector before starting the inference. But don't worry about make that. You just run below command. The facebank_vector will be saved in facebank folder.

python
make_facebank.py

3. Run inference

Now, every preprocessing is ended. you enter the image_path at infer.py. That's it!

python
infer.py