BinhuiXie/DCAN

Code release for "Domain Conditioned Adaptation Network" (AAAI 2020)

Domain Conditioned Adaptation Network

Paper

Domain Conditioned Adaptation Network (AAAI Conference on Artificial Intelligence, 2020)

If you find this code useful for your research, please cite our paper:

@inproceedings{Li20DCAN,
    title = {Domain Conditioned Adaptation Network},
    author = {Li, Shuang and Liu, Chi Harold and Lin, Qiuxia and Xie, Binhui and Ding, Zhengming and Huang, Gao and Tang, Jian},
    booktitle = {Thirty-Fourth AAAI Conference on Artificial Intelligence (AAAI-20)},    
    year = {2020}
}

Abstract

Tremendous research efforts have been made to thrive deep domain adaptation (DA) by seeking domain-invariant features. Most existing deep DA models only focus on aligning feature representations of task-specific layers across domains while integrating a totally shared convolutional architecture for source and target. However, we argue that such strongly-shared convolutional layers might be harmful for domain-specific feature learning when source and target data distribution differs to a large extent. In this paper, we relax a shared-convnets assumption made by previous DA methods and propose a Domain Conditioned Adaptation Network (DCAN), which aims to excite distinct convolutional channels with a domain conditioned channel attention mechanism. As a result, the critical low-level domain-dependent knowledge could be explored appropriately. As far as we know, this is the first work to explore the domain-wise convolutional channel activation for deep DA networks. Moreover, to effectively align high-level feature distributions across two domains, we further deploy domain conditioned feature correction blocks after task-specific layers, which will explicitly correct the domain discrepancy. Extensive experiments on three cross-domain benchmarks demonstrate the proposed approach outperforms existing methods by a large margin, especially on very tough cross-domain learning tasks.

Prerequisites

The code is implemented with Python(3.7) and Pytorch(1.2.0).

To install the required python packages, run

pip install -r requirements.txt

Datasets

Office-Home

Office-Home dataset can be found here.

DomainNet

DomainNet dataset can be found here.

Office-31

Office-31 dataset can be found here.

Pre-trained models

Pre-trained models can be downloaded here and put in <root_dir>/pretrained_models

Running the code

Office-Home

$ python train_dcan.py --gpu_id id --net 50 --output_path snapshot/ --data_set home --source_path data/list/home/Art_65.txt --target_path data/list/home/Clipart_65.txt --test_path data/list/home/Clipart_65.txt --task ac

DomainNet

$ python train_dcan.py --gpu_id id --net 50/101/152 --output_path snapshot/ --data_set domainnet --source_path /data/list/domainnet/clipart_train.txt --target_path data/list/domainnet/infograph_train.txt --test_path data/list/domainnet/infograph_test.txt --task ci

Office-31

$ python train_dcan.py --gpu_id id --net 50 --output_path snapshot/ --data_set office --source_path data/list/office/dslr_31.txt --target_path data/list/office/webcam_31.txt --test_path data/list/office/webcam_31.txt --task dw

Acknowledgements

This code is heavily borrowed from Xlearn and CDAN.

Contact

If you have any problem about our code, feel free to contact

• shuangli@bit.edu.cn

or describe your problem in Issues.