Hierarchical Filtering and Refinement Classification for Few-Shot Class-Incremental Learning

Li-Jun Zhao, Zhen-Duo Chen, Xin Luo, and Xin-Shun Xu

@article{
zhao2026hierarchical,
title={Hierarchical Filtering and Refinement Classification for Few-Shot Class-Incremental Learning},
author={Li-Jun Zhao and Zhen-Duo Chen and Xin Luo and Xin-Shun Xu},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2026},
url={https://openreview.net/forum?id=7MXra1JSh8},
note={J2C Certification}
}

Requirements

• PyTorch >= version 1.1
• tqdm

Data Preparation

We provide the source code on three benchmark datasets, i.e., CUB200, miniImageNet, and CIFAR100.

We follow CEC setting to use the same data index_list for training.
For CIFAR100, the dataset will be download automatically.
For miniImagenet and CUB200, you can download from here. Please unzip the downloaded file and change data_root to the location of the datasets:

$ tar -xvf miniimagenet.tar 
$ tar -xvzf CUB_200_2011.tgz

Scripts

Running the following shell scripts to train and evaluate the model with hyperparameters matching our paper.

Base training

mini_imagenet: run_base_miniImageNet.sh

cifar100: run_base_CIFAR100.sh

cub200: run_base_CUB200.sh

Inference

mini_imagenet: run_inc_prototype_miniImageNet.sh

cifar100: run_inc_prototype_CIFAR100.sh

cub200: run_inc_prototype_CUB200.sh

Acknowledgment

Our project references the codes in the following repos.

• CEC