EASE: Unsupervised Discriminant Subspace Learning for Transductive Few-Shot Learning (CVPR2022)
This repository is the official implementation of EASE: Unsupervised Discriminant Subspace Learning for Transductive Few-Shot Learning. This code is based on the Leveraging the Feature Distribution in Transfer-based Few-Shot Learning and Contrastive Laplacian Eigenmaps.
usage
Please download features from here and unzip it. This code can reproduce the performance in the paper, but something need to tune in different datasets like L2 normalization of prototypes and soft-labels (>1) for the computation of sinkhorn distance in the SIAMESE part.
If there is anything confusing you, please let me know.
References
Contrastive Laplacian Eigenmaps
Leveraging the Feature Distribution in Transfer-based Few-Shot Learning
Charting the Right Manifold: Manifold Mixup for Few-shot Learning
Manifold Mixup: Better Representations by Interpolating Hidden States
Sinkhorn Distances: Lightspeed Computation of Optimal Transport