Dual Consolidation for Pre-Trained Model-Based Domain-Incremental Learning

January 28, 2026 · View on GitHub

Da-Wei Zhou Zi-Wen Cai Han-Jia Ye Lijun Zhang De-Chuan Zhan

School of Artificial Intelligence, State Key Laboratory for Novel Software Technology, Nanjing University

The code repository for "Dual Consolidation for Pre-Trained Model-Based Domain-Incremental Learning" (CVPR 2025) in PyTorch. If you use any content of this repo for your work, please cite the following bib entry:

@inproceedings{zhou2025duct,
    title={Dual Consolidation for Pre-Trained Model-Based Domain-Incremental Learning},
    author={Da-Wei Zhou and Zi-Wen Cai and Han-Jia Ye and Lijun Zhang and De-Chuan Zhan},
    booktitle={CVPR},
    year={2025},
}

Updates

[02/2025] Code has been released.

[02/2025] Accepted to CVPR 2025.

[10/2024] arXiv paper has been released.

Introduction

Domain-Incremental Learning (DIL) involves the progressive adaptation of a model to new concepts across different domains. While recent advances in pre-trained models provide a solid foundation for DIL, learning new concepts often results in the catastrophic forgetting of pre-trained knowledge. Specifically, sequential model updates can overwrite both the representation and the classifier with knowledge from the latest domain. Thus, it is crucial to develop a representation and corresponding classifier that accommodate all seen domains throughout the learning process. To this end, we propose DUal ConsolidaTion (DUCT) to unify and consolidate historical knowledge at both the representation and classifier levels. By merging the backbone of different stages, we create a representation space suitable for multiple domains incrementally. The merged representation serves as a balanced intermediary that captures task-specific features from all seen domains. Additionally, to address the mismatch between consolidated embeddings and the classifier, we introduce an extra classifier consolidation process. Leveraging class-wise semantic information, we estimate the classifier weights of old domains within the latest embedding space. By merging historical and estimated classifiers, we align them with the consolidated embedding space, facilitating incremental classification. Extensive experimental results on four benchmark datasets demonstrate DUCT's state-of-the-art performance.

Requirements

Environment

You can create a conda environment and run the following command to install the dependencies.

conda install --file requirements.txt

Dataset

There are 4 datasets involved in the paper, CDDB, CORe50, DomainNet and Office-Home respectively. Follow the two-step guideline to prepare them for the reproduction.

Download the datasets mannually according the recommended.
- CDDB: You can access the dataset at this link.
- CORe50: You can access the dataset at this link. Please download the core50_imgs.npz file.
- DomainNet: You can access the dataset at this link. Please download the cleaned version file.
- Office-Home: You can access the dataset at this link.

Check if the dataset has been downloaded properly. The dataset directory is expected to have the following structure:

CDDB
├── biggan
├── crn 
├── ...
├── wild

CORe50
├── s1
├── ...
├── s11
├── labels.pkl
├── LUP.pkl
├── paths.pkl

DomainNet
├── clipart
├── ...
├── sketch
├── clipart_test.txt
├── clipart_train.txt
├── ...

OfficeHome
├── Art
├── Clipart
├── Product
├── Real_World

Specify the dataset path in the config file under the configs directory.
```
"data_path": "$Your/Dataset/Path/Here$"
```

:warning: A note about DomainNet

In the 'cleaned' version of DomainNet dataset, we identified a label mismatch in the painting domain:

Issue: the painting_train.txt file doesn't include the t-shirt category with label 327, though it exists in the test set painting_test.txt. This discrepancy could cause errors during evaluation.
Fix: In our experiments, we resolved this by moving the following line from the test file to the train file: painting/t-shirt/painting_328_000001.jpg 327.

Running scripts

python main.py --config configs/Template_domainnet_duct.json