DDA-MLIC

This repo is the official PyTorch implementation of "Discriminator-free Unsupervised Domain Adaptation for Multi-label Image Classification" in WACV 2024.

Citation

@inproceedings{singh2023discriminatorfree,
  title={Discriminator-free Unsupervised Domain Adaptation for Multi-label Image Classification},
  author={Singh, Indel Pal and Ghorbel, Enjie and Kacem, Anis and Rathinam, Arunkumar and Aouada, Djamila},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  year={2024}
}

Datasets

1. AID
2. UCM (Labels)
3. DFC
4. PASCAL VOC 2007
5. Clipart1k
6. Cityscapes
7. Foggycityscapes

Pre-trained models

We provide a collection of models trained with the proposed GMM-based discrepancy on various multi-label domain adaptation datasets.

Source	Target	mAP
AID	UCM	63.2
UCM	AID	54.9
AID	DFC	62.1
UCM	DFC	70.6
VOC	Clipart	61.4
Clipart	VOC	77.0
Cityscapes	Foggycityscapes	62.3

Installation

Create virtual environment

$ python3 -m venv dda_mlic

Activate your virtual environment

$ source dda_mlic/bin/activate

Upgrade pip to the latest version

$ pip install --upgrade pip

Install compatible CUDA and pytorch versions.

$ pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html

Install other required packages from requirements.txt

$ pip install -r requirements.txt

Evaluation

Syntax: Source → Target

$ python main.py --phase test -s <name_of_source_dataset> -t <name_of_target_dataset> -s-dir <path_to_source_dataset_dir> -t-dir <path_to_target_dataset_dir> --model-path <path_to_pretrained_weights>

Example: AID → UCM

$ python main.py --phase test -s AID -t UCM -s-dir datasets/AID -t-dir datasets/UCM --model-path models/aid2ucm_best_63-2.pth

Training

Download the imagenet pretrained weights for TResNetM.

Syntax: Source → Target

$ python main.py -s <name_of_source_dataset> -t <name_of_target_dataset> -s-dir <path_to_source_dataset_dir> -t-dir <path_to_target_dataset_dir> --model-path <path_to_imagenet_pretrained_weights>

Example: AID → UCM

$ python main.py -s AID -t UCM -s-dir datasets/AID -t-dir datasets/UCM --model-path models/tresnet_m_miil_21k.pth

Acknowledgement

We create our code based on the following repositories:

Transfer Learning Library

TResNet/ASL

ImageNet21k

Thanks to the authors.