Skip to content

Latest commit

 

History

History
64 lines (49 loc) · 2.67 KB

README.md

File metadata and controls

64 lines (49 loc) · 2.67 KB
Raw

Project

The project includes two models:

  • Face detection. Detection is done by outputing a heatmap of faces. Using this heatmap bounding boxes are created.

  • Face recognition. Detected faces are cropped using the bounding boxes and encoded into a vector of 126 values. This vector is then compared to the vectors of faces stored in data/infer_faces.

❗ Performance of the detection model is reasonable but I'm still strugling with recognition model.

Demo

Demo

Project structure

    .
    ├── checkpoints/                    # dir for model weights
    ├── data/                           # dir for data storing and generation 
    │   ├── infer_faces/                # directory for storing pictures of faces to detect during inference
    │   ├── other/                      # other none code files video/gifs etc.
    │   ├── imdb_face.py                # imdb dataset utils (used for recognition training)
    │   └── wider_face.py               # wider face dataset utils (used for detection training)
    ├── models/                         # dir for model definitions
    │   ├── detection.py                # -
    │   ├── recognition.py              # -
    │   └── face_align.py               # not implemented
    ├── infer.py                        # -
    ├── train_detection.py              # -
    ├── train_recognition.py            # -
    ├── utils.py                        # -
    └── README.md                       # -

Installation and dependencies

To run this project the following libraries are required: tensorflow, numpy, opencv, pillow, pandas.
Run this on your own env or create a new one using environment.yml

git clone https://github.com/RaidasGrisk/tf2-face-detection-recognition.git  
cd tf2-face-detection-recognition  

conda env create -f environment.yml  
conda activate tf2-face-detection-recognition

Inference

Place images of people's faces inside data/infer_faces (see examples inside this dir). These images will be encoded and compared with faces detected during inference.

python infer.py

Training

For detection model explore train_detection.py and data/wider_face.py files. Fore recognition model explore train_recognition.py data/imdb_face.py files.

To re-train or continue training the models you would need datasets which are not included as part of this git.

Detection model is trained on wider face dataset.
Recognition model is trained on imdb face dataset.