Pair-wise-Similarity-module

Code release for the paper "Learning Calibrated Class Centers for Few-shot Classification by Pair-wise Similarity"

Requirements

• python=3.6
• PyTorch=1.2+
• torchvision=0.4.2
• pillow=6.2.1
• numpy=1.18.1
• h5py=1.10.2

Dataset

• CUB-200-2011
  Change directory to ./filelists/CUB
  run source ./download_CUB.sh
  

Train

• method: relationnet | relationnet_PSM | protonet | protonet_PSM.
• n_shot: number of labeled data in each class （1, 5）.
• train_aug: perform data augmentation or not during training.

python ./train.py --dataset CUB --model Conv4 --method protonet --n_shot 1
python ./train.py --dataset CUB --model Conv4 --method protonet_PSM --n_shot 1

Save features

python ./save_features.py --dataset CUB --model Conv4 --method protonet --n_shot 1
python ./save_features.py --dataset CUB --model Conv4 --method protonet_PSM --n_shot 1

Test

python ./test.py --dataset CUB --model Conv4 --method protonet --n_shot 1
python ./test.py --dataset CUB --model Conv4 --method protonet_PSM --n_shot 1

References

Our code is based on Chen's contribution. Specifically, except for our core design, protonet_PSM and relationnet_PSM, everything else （e.g. backbone, dataset, relation network, evaluation standards, hyper-parameters）are built on and integrated in https://github.com/wyharveychen/CloserLookFewShot.

Contact

Thanks for your attention! If you have any suggestion or question, you can leave a message here or contact us directly:

guoyurong@bupt.edu.cn