Awesome
Kernel Relative-prototype Spectral Filtering for Few-shot Learning
This repository contains the code for the paper: Kernel Relative-prototype Spectral Filtering for Few-shot Learning
Citation:
@inproceedings{zhang2022kernel,
title={Kernel Relative-prototype Spectral Filtering for Few-shot Learning},
author={Zhang, Tao and Huang, Wu},
booktitle={European Conference on Computer Vision},
year={2022}
organization={Springer}
}
Dependencies:
- Python 3.9.7
- PyTorch 1.11.0
- qpth 0.0.15
- tqdm
Dataset:
miniImageNet, tieredImageNet, CIFAR-FS
Training:
-
Train shrinkage classifier using ResNet-12 on 5-way mini-ImageNet 5-shot/1-shot benchmark:
python train.py --gpu 0,1,2,3 --save-path "./experiments/miniImageNet_shrinkage" --train-shot 15 --train-query 10\ --head shrinkage --network ResNet --dataset miniImageNet --eps 0.1 -
Train shrinkage classifier using Conv-4 on 5-way mini-ImageNet 5-shot/1-shot benchmark:
python train.py --gpu 0 --save-path "./experiments/miniImageNet_shrinkage" --train-shot 15 --train-query 10\ --head shrinkage --network ProtoNet --dataset miniImageNet -
Train shrinkage classifier using ResNet-12 on 5-way tieredImageNet 5-shot/1-shot benchmark:
python train.py --gpu 0,1,2,3 --save-path "./experiments/tieredImageNet_shrinkage" --train-shot 10 --train-query 15\ --head shrinkage --network ResNet --dataset tieredImageNet -
Train shrinkage classifier using ResNet-12 on 5-way CIFAR-FS 1-shot benchmark:
python train.py --gpu 0 --save-path "./experiments/CIFAR_FS_shrinkage" --train-shot 2 --train-query 20\ --head shrinkage --network ResNet --dataset CIFAR_FS -
Train shrinkage classifier using ResNet-12 on 5-way CIFAR-FS 5-shot benchmark:
python train.py --gpu 0 --save-path "./experiments/CIFAR_FS_shrinkage" --train-shot 15 --train-query 10\ --head shrinkage --network ResNet --dataset CIFAR_FS
Testing:
- Test shrinkage classifier with ResNet-12 on 5-way miniImageNet 5-shot benchmark:
python test.py --gpu 0 --load ./experiments/miniImageNet_shrinkage/best_model.pth --episode 1000 \
--way 5 --shot 5 --query 15 --head shrinkage --network ResNet --dataset miniImageNet
- Train shrinkage classifier with ResNet-12 on 5-way miniImageNet 1-shot benchmark:
python test.py --gpu 0 --load ./experiments/miniImageNet_shrinkage/best_model.pth --episode 1000 \
--way 5 --shot 1 --query 15 --head shrinkage --network ResNet --dataset miniImageNet
Acknowledgement:
This code is based on the codebases: