Pytorch implementation of one shot learning for facial embedding . The model is trained on dlib aligned pinterest faces (https://www.kaggle.com/frules11/pins-face-recognition/)
step 1 : Download the dataset (Above link)
step 2 : generate csv dataset using csvGenerator notebook
step 3. Explore and train the model using one shot learning notebook
1. Add more training data
2.Maximizing the batch size
3.Maximizing the marigins
4. Use ResNet 50 or more complex Networks as embedders
5. Run for more epochs
6. Semihard and online hard sampling