Prototypical Networks

ProtoNets learns a metric space for solving few-shot classification by computing distances between the query image and a prototype representation of the class.

Requirements

• PyTorch
• Numpy
• Scipy
• OpenCV
• Matplotlib
• Tqdm

Usage

1. Download the train and evaluation datasets.
2. Unzip the downloaded zip files in a folder called omniglot/.
3. Run the code cells in ProtoNets-Omniglot.ipynb.
4. Alternatively, you can run this repository in colab

Credits

• Prototypical Networks paper
• Author's Implementation https://github.com/jakesnell/prototypical-networks