A few-shot classification algorithm: Charting the Right Manifold: Manifold Mixup for Few-shot Learning

Our code is built upon the code base of A Closer Look at Few-shot Classification and Manifold Mixup: Better Representations by Interpolating Hidden States

Dataset: mini-ImageNet, CIFAR-FS, CUB

Donwloading the dataset:

CUB

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

CIFAR-FS

• Change directory to filelists/cifar/
• run 'source ./download_cifar.sh'

miniImagenet

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

Training

DATASETNAME: miniImagenet/cifar/CUB

METHODNAME: S2M2_R/rotation/manifold_mixup

For CIFAR-10

python train_cifar.py --method [METHODNAME] --model WideResNet28_10

For miniImagenet/CUB

python train.py --dataset [DATASETNAME] --method [METHODNAME] --model WideResNet28_10

Fetching WideResNet_28_10 model checkpoints for evaluation

Create an empty 'checkpoints' directory inside 'S2M2'

The model for each dataset can be downloaded from this link - https://drive.google.com/open?id=1S-t56H8YWzMn3sjemBcwMtGuuUxZnvb_

Move the tar files for each dataset into 'checkpoints' folder and untar it. E.g. tar -xvzf cifar_model.tar.gz

Saving the features of a checkpoint for checkpoint evalution

python save_features.py --dataset [DATASETNAME] --method [METHODNAME] --model WideResNet28_10

Fetching novel class features for evaluation

Create an empty 'features' directory inside 'S2M2'

Features can be be directly downloaded at this link 'https://drive.google.com/open?id=1JtA7p3sDPksvBmOsJuR4EHw9zRHnKurj' for easy evaluation without the need to download datasets and models. Move the tar files for each dataset into 'features' folder and untar it.

Evaluating the few-shot performance

python test.py --dataset [DATASETNAME] --method [METHODNAME] --model WideResNet28_10 --n_shot [1/5]

Comparison with prior/current state-of-the-art methods on mini-ImageNet, CUB and CIFAR-FS dataset. Note: We implemented LEO on CUB dataset. Other numbers are reported directly from the paper.

A Closer Look at Few-shot Classification

Meta-Learning with Latent Embedding Optimization

Meta Learning with Differentiable Convex Optimization

Manifold Mixup: Better Representations by Interpolating Hidden States