OSFI-by-FineTuning
Official implementation for Open-set Face Identification on Few-shot Gallery by Fine-Tuning.
Published on 2022 International Conference on Pattern Recognition (Paper).
ICPR2022 Presentation
Presentation_resized.mp4
Requirements
- Pytorch 1.7.1
- Torchvision 0.8.2
Setup
Download Pretrained Weights:
Download Dataset:
Above is the CASIA-WebFace dataset. The images are already cropped using MTCNN by timesler.
To use your own face dataset, you can simply change the data_config in config.py.
The face dataset must have the structure ROOT/SUBJECT_NAME/image.jpg.
After downloading, change the dataset_config and encoder_config in config.py accordingly.
Usage
After the setup is done, simply run:
python main.py --dataset='CASIA' --encoder='VGG19' --classifier_init='WI' --finetune_layers='BN'
For further information on the arguments, please refer to our paper.