Knowledge-Guided Semantics Adjustment (KGSA)

This is the code for the ICDM 2022 Paper: Knowledge-Guided Semantics Adjustment for Improved Few-Shot Classification.

Requirements

• Python3
• Pytorch==1.7.1

How to Use

Datasets:

1. miniImageNet
2. tieredImageNet

Rename the folder for the miniImageNet dataset as miniImagenet and rename the folder for the tieredImageNet dataset astiered_imagenet. The root_path in the dataset configuration is the parent folder of a dataset folder.

To pre-train the embedding network on the miniImageNet and the tieredImageNet, run

python train_classifier.py --config configs/train_classifier_mini.yaml
python train_classifier.py --config configs/train_classifier_tiered.yaml

To train the KGSA on the miniImageNet and the tieredImageNet, run

python train_kgsa.py --config configs/train_kgsa_mini.yaml
python train_kgsa.py --config configs/train_kgsa_tiered.yaml

Citation

@inproceedings{zheng2022knowledge,
  title={Knowledge-Guided Semantics Adjustment for Improved Few-Shot Classification},
  author={Zheng, Guangtao and Zhang, Aidong},
  booktitle={2022 IEEE International Conference on Data Mining (ICDM)},
  pages={1347--1352},
  year={2022},
  organization={IEEE}
}