W251_Project

Final Project for W251 MIDS Program

1. Download vggFace2 Dataset

Download dataset with 500 faces from (https://www.kaggle.com/greatgamedota/vggface2-test). The dataset is stored in W251_Project/data/test_no_mask folder

2. Dataset Augmentation with masks

Install packages in conda environment given in W251_project/conda_env.sh

python mask_the_face.py --inpath data/test_no_mask --outpath data/test_with_mask --mask_type <type-of-mask> --verbose --write_original_image

3. Facenet training without masks

Run jupyter notebook

jupyter notebook --allow-root

Run finetune_no_augmentation.ipynb

The trained model is saved as /data/model_no_mask.pt

4. Facenet training with masks

Run finetune_with_augmentation.ipynb

The trained model is saved as /data/model_with_mask.pt

The training with and without the augmentation are done to demonstrate improvements with the augmentation.

5. Masked face Inference on Jetson

The trt versions of models are stored in W251_Project/models/trt_engines.

• Run pip3 install -e . (from inference folder) to generate python package
• Run setup.sh to install pycuda
• anchors are in W251_Project/data/test_anchors folder and test data in W251_Project/data/test_no_mask
• Run sudo python3 test_mtcnn.py (inference folder)