CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning

July 14, 2025 · View on GitHub

This repository contains the official PyTorch implementation of CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning, accepted at CVPR 2025.

📄 Paper: CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning

Overview

CL-LoRA introduces a novel approach for class-incremental learning without rehearsal, leveraging low-rank adaptation techniques to efficiently learn new classes while preserving knowledge of previously learned tasks.

Getting Started

Environments

python==3.9.4
torch==2.0.1
torchvision==0.15.2
timm==0.6.12
numpy==1.25.2
scikit-learn==1.2.0
cudatoolkit==11.1

Training

To train the model, navigate to the main directory and run:

python main.py <json_config_path>

Supported Datasets

The following datasets are supported with pre-configured JSON files:

CIFAR-100

python main.py ./exps/cifar.json

ImageNet-R

python main.py ./exps/inr.json

ImageNet-A

python main.py ./exps/ina.json

VTAB

python main.py ./exps/vtab.json

Configuration

The JSON configuration files contain experiment setup and hyperparameters. Key CL-LoRA specific parameters include:

general_pos: Position indices for task-shared LoRA adapters
specific_pos: Position indices for task-specific LoRA adapters
msa: Multi-Head Self-Attention adaptation settings for Query (Q), Key (K), and Value (V)
- 1: Apply adaptation
- 0: No adaptation

Example Configuration

{
  "msa": [1, 0, 1],
  "general_pos": [0, 1, 2, 3, 4, 5],
  "specific_pos": [6, 7, 8, 9, 10, 11]
}

This configuration:

Adapts Q and V in Multi-Head Self-Attention (msa: [1, 0, 1])
Uses task-shared LoRA in the first 6 ViT blocks (general_pos)
Uses task-specific LoRA in the last 6 ViT blocks (specific_pos)

Citation

If you find this work useful in your research, please cite:

@article{He_2025_CVPR,
    author    = {He, Jiangpeng and Duan, Zhihao and Zhu, Fengqing},
    title     = {CL-LoRA: Continual Low-Rank Adaptation for Rehearsal-Free Class-Incremental Learning},
    journal = {Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
    month     = {June},
    year      = {2025},
    pages     = {30534-30544}
}