SLK-few-shot

Clustering for Transductive Few-shot Learning

This repository contains the code for our paper entitled:

Transductive Few-Shot Learning: Clustering is All You Need?
Imtiaz Masud Ziko, Malik Boudiaf, Jose Dolz, Eric Granger and Ismail Ben Ayed, ArXiv 2021

Introduction

We adapt several clustering methods to transductive inference in few-shot learning tasks. The clustering part works on a feature extractor initially trained over the base-class data. Using standard training on the base classes, without resorting to complex meta-learning and episodic-training strategies, our regularized-clustering approaches outperform state-of-the-art few-shot methods by significant margins, across various models, settings and data sets. Surprisingly, we found that even standard clustering procedures (e.g., K-means), which correspond to particular, non-regularized cases of our general model, already achieve competitive performances in comparison to the state-of-the-art in transductive few-shot learning. These surprising results point to the limitations of the current few-shot benchmarks, and question the viability of a large body of convoluted few-shot learning techniques in the recent literature.

Usage

1. Dependencies

• Python 3.6+
• Pytorch 1.0+

2. Datasets

2.1 Mini-ImageNet

You can download the dataset from here

2.2 Tiered-ImageNet

You can download the Tiered-ImageNet from here. After downloading and unziping this dataset run the following script to generate split files.

python src/utils/tieredImagenet.py --data path-to-tiered --split split/tiered/

2.3 CUB

Download and unpack the CUB 200-2011 from here After downloading and unziping this dataset run the following script to generate split files.

python src/utils/cub.py --data path-to-cub --split split/cub/

3 Train and Test

You can download our pretrained network models on base classes by running:

cd ./src
python download_models.py

Alternatively to train the network on the base classes from scratch remove the "--evaluate " options in the following script. The scripts to test SLK-shot:

sh run_SLK.sh

Results

We get the following results in different few-shot benchmarks:

On mini-ImageNet

With WRN network:

On tiered-ImageNet

With WRN network:

On CUB

With ResNet-18 network