Head Pose Estimation using FSA Net

Scientific research 2022 about Head Pose Estimation using modified FSA Net

Demo

Video file or a camera index can be provided to demo script. If no argument is provided, default camera index is used.

Video file usage

For any video format that OpenCV supported (mp4, avi etc.):

python3 demo.py --video /path/to/video.mp4

Camera usage

python3 demo.py --cam 0

Training and testing

1. Original FSANet:

For training, checkout the notebook: src/2-Train Model.ipynb.

For testing, checkout the notebook: src/2-Test Model.ipynb.

I make two Python files from those notebooks named src/train_fsa.py and src/test_fsa.py in case you want to run locally instead of using .ipynb files on Google Colab.

2. FSANet combined with Triplet Network architecture:

Basically, everything is the same as in part 1, but please use the modified files in folder src_triplet/ instead of src/

Dataset

For model training and testing, you can download the preprocessed dataset from author's official git repository and place them inside the data/ directory. Your dataset hierarchy should look like this:

data/
  type1/
    test/
      AFLW2000.npz
    train/
      AFW.npz
      AFW_Flip.npz
      HELEN.npz
      HELEN_Flip.npz
      IBUG.npz
      IBUG_Flip.npz
      LFPW.npz
      LFPW_Flip.npz

Acknowledgements

This work is based on:

• The FSA Net repo and paper of Yang et al.
• A third-party Pytorch implementation github repo (This is where all the files are from, some of them are modified for training FSANet with the Triplet Network architecture)