This is the code for Achieving Better Kinship Recognition Through Better Baseline (FG2020 workshop).

1. mxnet
2. insightface
3. gluonfr
4. OpenCV
5. matplotlib
6. tqdm

There are two scripts that can be run: verification.py and train.py. The first provides nessesary functions to prepare (as described in the section III-A of the paper) the dataset and run a validation and plot a ROC curve. The second one provides necessary code to reproduce our training procedure (the results can vary slightly).

We provide already prepared data (re-detected and re-aligned) that can be found in train-faces-det and val-faces-det folders for training and validation images respectfully.

You can downloads models through Git LFS while cloning the repository. Alternatively, you can download them from this link.

If you'd like to cite the training pipeline or results from the paper, use this citation:

@inproceedings{shadrikov2020fitw,
  title={Achieving Better Kinship Recognition Through Better Baseline},
  author={Shadrikov, Andrei},
  booktitle={2020 15th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2020)(FG)},
  pages={872--876},
  organization={IEEE Computer Society}
}

For the Families in the Wild dataset, please see it's homepage.