Regression Networks for Meta-Learning Few-Shot Classification

This repository contains the reference source code for our paper Regression Networks for Meta-Learning Few-Shot Classification.

Part of this work has been presented at the ICML 2020 Workshop on Automated Machine Learning.

Citation

If you find our code/paper useful, please consider citing our work using the bibtex:

@article{devos2020RegressionNet,
    title="{Regression Networks for Meta-Learning Few-Shot Classification}",
    author={Arnout Devos and Matthias Grossglauser},
    journal={7th ICML Workshop on Automated Machine Learning},
    year={2020}
}

Enviroment

• Python == 3.6.10
• PyTorch == 1.5.1

Getting started

Make sure that, after downloading a dataset as instucted below, you set data_dir['DATASETNAME'] in configs.py to the folder path.

mini-ImageNet

• Change directory to ./filelists/miniImagenet
• run source ./download_miniImagenet.sh

(WARNING: This would download the 155GB ImageNet dataset. You can comment out lines 5-6 in download_miniImagenet.sh if you already have this dataset.)

CUB

• Change directory to ./filelists/CUB
• run source ./download_CUB.sh

mini-ImageNet -> CUB (cross)

• Finish preparation for CUB and mini-ImageNet and you are done!

Train

Run python ./train.py --dataset [DATASETNAME] --model [BACKBONENAME] --method [METHODNAME] [--OPTIONARG]

For example, run python ./train.py --dataset miniImagenet --model Conv4 --method regressionnet --train_aug --lamb 0.01
Commands below follow this example, and please refer to io_utils.py for additional options.

Save features

Save the extracted feature before the classifaction layer to increase test speed. This is not applicable to MAML, but is required for other methods. Run python ./save_features.py --dataset miniImagenet --model Conv4 --method regressionnet --train_aug --train_n_shot 5 --test_n_shot 5

Test

Run python ./test.py --dataset miniImagenet --model Conv4 --method regressionnet --train_aug --train_n_shot 5 --test_n_shot 5

Results

• The test results will be recorded in ./record/results.txt

References

This testbed builds mostly upon the repository below.

A Closer Look at Few-shot Classification:
https://github.com/wyharveychen/CloserLookFewShot