bergsalex / roux-siamese-few-shot

Siamese Network implementation using Pytorch for Few-Shot Learning

pytorch-siamese

This is a port of chainer-siamese and a fork of pytorch-siamese. It implements additional training datasets, as well as few-shot prediction code for the siamese network.

Install

This installation requires cuda to be installed.

$ poetry install

or

$ python3 -m venv env
$ source env/bin/activate
$ pip install -r requirements.txt

Run

$ env/bin/python train_mnist.py --epoch 10
$ env/bin/python train_omniglot.py --epoch 10
$ env/bin/python train_omniglot_by_alphabet.py --epoch 10

This dumps for every epoch the current state and creates a result.png. Model state is saved in a new directory unique to each run.