Siamese Neural Networks for Few Shot Learning
The problem has been inspired by Fellowship.AI, demo challenge
-
The model has been trained using tensforflow backend in Keras.
-
It tries to solve the problem of image verification when the quantity of data available for training Deep Learning models is less.
-
The model has been implemented to solve the problem based on the paper by Gregory et. al Siamese Neural Networks for One-Shot Image Recognition.
- The ideas from the paper have been used to extend the model for few shot learning.
-
Omniglot, dataset has been used for training the model
- The dataset has 1623 character classes, each with 20 examples.
- The model tries to build a few shot classifier using the ides presented in the paper by Gregory et. al for the dataset.