Transfer learning based few-shot classification using optimal transport mapping from preprocessed latent space of backbone neural network

This is the official repository for the Transfer learning based few-shot classification using optimal transport mapping from preprocessed latent space of backbone neural network papers introducting the Latent Space Transform algorithm for preproccesing backbone-extracted feature vectors to resemble Gaussian-like distributions.

https://arxiv.org/abs/2102.05176

Training

We use the same backbone network and training strategies as 'S2M2_R'. For more information about the backbone training, please refer to https://github.com/nupurkmr9/S2M2_fewshot.

Pre-trained models

The trained models at miniImageNet, CIFAR-FS and CUB can be found on the following link: https://drive.google.com/file/d/1-DM1GSVpjQhjEbHvNdDq9oKc-DnNlt8i/view?usp=sharing.

The archive needs to be extracted in /checkpoints directory in the repository (you need to make the directory by yourself).

Please adjust the paths accordingly before running the model.

Execute the model by running the following command:

python3 LST_run.py