FSCIL-ASP Official Implementation

This codebase contains the official Python implementation of Few-shot Class Incremental Learning with Attention-Aware Self-Adaptive Prompt (ECCV2024)

Introduction

ASP is a novel few-shot class incremental learning (FSCIL) algorithm which utilizes prompt tuning with a Vision Transformer backbone. ASP encourages task-invariant prompts to capture shared knowledge by reducing specific information from the attention aspect. Additionally, self-adaptive task-specific prompts in ASP provide specific information and transfer knowledge from old classes to new classes with an Information Bottleneck learning objective.

Performance

ASP consistently outperforms traditional FSCIL works using ResNet, multi-modal FSCIL works using CLIP, and prompt-based CIL works using ViT.

Instructions on running ASP

Environment setup

Clone this GitHub repository and run:

pip install -r requirements.txt

Dataset preparation

Download the dataset and put them in folder ./data

• CIFAR100: will be automatically downloaded by the code.
• CUB200: Google Drive: link
• ImageNet-R: Google Drive: link

Run experiment

Experiment on CIFAR100 dataset:

python main.py --config=./exps/cifar.json

Experiment on CUB200 dataset:

python main.py --config=./exps/cub.json

Experiment on ImageNet-R dataset:

python main.py --config=./exps/inr.json

Citation

@article{liu2024few,
  title={Few-Shot Class Incremental Learning with Attention-Aware Self-Adaptive Prompt},
  author={Liu, Chenxi and Wang, Zhenyi and Xiong, Tianyi and Chen, Ruibo and Wu, Yihan and Guo, Junfeng and Huang, Heng},
  journal={arXiv preprint arXiv:2403.09857},
  year={2024}
}