jiaxinchen666 / Meta2Learning

A pytorch implementation of Meta$^2$ Learning.

Prerequisites

• Python >= 3.5
• Pytorch >= 1.3 and torchvision (https://pytorch.org/)
• You can use the requirements.txt file we provide to setup the environment via Anaconda.

conda create --name py36 python=3.6
conda install pytorch torchvision -c pytorch
pip3 install -r requirements.txt

Install

Clone this repository:

git clone https://github.com/jiaxinchen666/Meta2Learning.git
cd CrossDomainFewShot

Datasets

• Set DATASET_NAME to: cars, cub, miniImagenet, flowers, pets, fungi, or plantae.

cd filelists
python process.py DATASET_NAME
cd ..

For datasets 'vegetable', 'food', and 'butterfly', download vegetable, butterfly, food and extract them into ./filelists/DATASET_NAME

cd filelists
python write_DATASET_filelist.py
cd ..

Feature encoder pre-training

We adopt baseline++ for MatchingNet, and baseline from CloserLookFewShot for other metric-based frameworks.

Download the pre-trained feature encoders.

cd output/checkpoints
python download_encoder.py
cd ../..

Training

--dataset A DATASET LIST CONSISTING OF TRAINING DOMAINS AND TEST DOMAIN

--n_shot 1/5, --method 'protonet'/'relationnet'/'gnnnet'/'matchingnet', --domain_specific 'True'/'False'(only for M2L)

Baseline

python train_baseline.py --n_shot 5 --testset DATASET_NAME --method METHOD --n_query 16 --mode 'onlytrain';

LFT (Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation)

python train.py --n_shot 5 --testset DATASET_NAME --method METHOD --n_query 16 --mode 'onlytrain';

M2L

python train_ours.py --n_shot 5 --testset DATASET_NAME --method METHOD --n_query 16 --mode 'onlytrain' --domain_specific 'True' --lr '0.1';

Evaluation

Baseline

python train_baseline.py --n_shot 5 --testset DATASET_NAME --method METHOD --n_query 16 --mode 'onlytest';

LFT (Cross-Domain Few-Shot Classification via Learned Feature-Wise Transformation)

python train.py --n_shot 5 --testset DATASET_NAME --method METHOD --n_query 16 --mode 'onlytest';

M2L

python train_ours.py --n_shot 5 --testset DATASET_NAME --method METHOD --n_query 16 --mode 'onlytest' --domain_specific 'True' --lr '0.1';

M2L (Further adaptation)

python finetune_test.py --n_shot 5 --testset DATASET_NAME --method METHOD --n_query 16; 

Note

• This code is built upon the implementation from CrossDomainFewShot and CloserLookFewShot.
• The dataset, model, and code are for non-commercial research purposes only.