Select, Label, and Mix (SLM)

The repository contains the codes for the paper "Select, Label, and Mix: Learning Discriminative Invariant Feature Representations for Partial Domain Adaptation" part of Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2023.

Aadarsh Sahoo¹, Rameswar Panda¹, Rogerio Feris¹, Kate Saenko^1,2, Abir Das³

¹ MIT-IBM Watson AI Lab, ² Boston University, ³ IIT Kharagpur

[Paper] [Project Page]

Preparing the Environment

Conda

Please use the slm_environment.yml file to create the conda environment SLM as:

conda env create -f slm_environment.yml

Pip

Please use the requirements.txt file to install all the required dependencies as:

pip install -r requirements.txt

Data Directory Structure

All the datasets should be stored in the folder ./data following the convention ./data/<dataset_name>/<domain_names>. E.g. for Office31 the structure would be as follows:

    .
    ├── ...
    ├── data
    │   ├── Office31
    │   │    ├── amazon
    │   │    ├── webcam
    │   │    ├── dslr
    │   └── ...
    └── ...

For using datasets stored in some other directories, please update the path to the data accordingly in the txt files inside the folder ./data_labels.

The official download links for the datasets used for this paper are:

Office31: https://people.eecs.berkeley.edu/~jhoffman/domainadapt/#datasets_code

OfficeHome: http://hemanthdv.org/OfficeHome-Dataset/

ImageNet-Caltech: http://www.image-net.org/, http://www.vision.caltech.edu/Image_Datasets/Caltech256/

VisDA-2017: http://ai.bu.edu/visda-2017/#download

Training SLM

Here is a sample and recomended command to train SLM for the transfer task of Amazon -> Webcam from Office31 dataset:

CUDA_VISIBLE_DEVICES=0 python main.py --manual_seed 1 --dataset_name Office31 --src_dataset amazon --tgt_dataset webcam  --batch_size 64 --model_root ./checkpoints_a31_w10 --save_in_steps 500 --log_in_steps 10 --eval_in_steps 10 --model_name resnet50 --classifier_name resnet50 --source_images_path ./data_labels/Office31/amazon_31_list.txt --target_images_path ./data_labels/Office31/webcam_10_list.txt --pseudo_threshold 0.3 --warmstart_models True --num_iter_adapt 10000 --num_iter_warmstart 5000 --learning_rate 0.0005 --learning_rate_ws 0.001

For detailed description regarding the arguments, use:

python main.py --help

Citing SLM

If you use codes in this repository, consider citing SLM. Thanks!

@inproceedings{sahoo2023select,
  title={Select, label, and mix: Learning discriminative invariant feature representations for partial domain adaptation},
  author={Sahoo, Aadarsh and Panda, Rameswar and Feris, Rogerio and Saenko, Kate and Das, Abir},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={4210--4219},
  year={2023}
}