BiMC

This is the official implementation of paper Enhancing Few-Shot Class-Incremental Learning via Training-Free Bi-Level Modality Calibration (CVPR 2025).

Abstract

Few-shot Class-Incremental Learning (FSCIL) challenges models to adapt to new classes with limited samples, presenting greater difficulties than traditional class-incremental learning. While existing approaches rely heavily on visual models and require additional training during base or incremental phases, we propose a training-free framework that leverages pre-trained visual-language models like CLIP. At the core of our approach is a novel Bi-level Modality Calibration (BiMC) strategy. Our framework initially performs intra-modal calibration, combining LLM-generated fine-grained category descriptions with visual prototypes from the base session to achieve precise classifier estimation. This is further complemented by inter-modal calibration that fuses pre-trained linguistic knowledge with task-specific visual priors to mitigate modality-specific biases. To enhance prediction robustness, we introduce additional metrics and strategies that maximize the utilization of limited data. Extensive experimental results demonstrate that our approach significantly outperforms existing methods.

Installation

Dataset

Please follow CEC to download mini-ImageNet, CUB-200 and CIFAR-100.

Requirement

• torch==1.13.1
• torchvision==0.14.1
• yacs==0.1.8
• tqdm==4.66.1
• ftfy==6.1.1
• regex==2023.10.3
• scikit-learn==1.3.2

Experiments

First, remember to modify the data path ROOT in the dataset configuration file.

# CIFAR BIMC
python main.py --data_cfg ./configs/datasets/cifar100.yaml --train_cfg ./configs/trainers/bimc.yaml

# CIFAR BIMC_Ensemble
python main.py --data_cfg ./configs/datasets/cifar100.yaml --train_cfg ./configs/trainers/bimc_ensemble.yaml

# MiniImagenet BIMC
python main.py --data_cfg ./configs/datasets/miniimagenet.yaml --train_cfg ./configs/trainers/bimc.yaml

# MiniImagenet BIMC_Ensemble
python main.py --data_cfg ./configs/datasets/miniimagenet.yaml --train_cfg ./configs/trainers/bimc_ensemble.yaml

# CUB200 BIMC
python main.py --data_cfg ./configs/datasets/cub200.yaml --train_cfg ./configs/trainers/bimc.yaml

# CUB200 BIMC_Ensemble
python main.py --data_cfg ./configs/datasets/cub200.yaml --train_cfg ./configs/trainers/bimc_ensemble.yaml

Acknowledgment

In this repository, we build our code based on the following excellent open-source projects. We sincerely thank all the authors for sharing their great work:

• LP-DiF
• TEEN
• FeCAM
• CuPL
• AdaptCLIPZS