uniot-for-unida's Introduction

[NeurIPS 2022 Spotlight] Unified Optimal Transport Framework for Universal Domain Adaptation

Code release for Unified Optimal Transport Framework for Universal Domain Adaptation (NeurIPS 2022 Spotlight).

Requirements

Python 3.7+
PyTorch 1.8.0
GPU Memory 12 GB

To install requirements:

pip install -r requirements.txt

Preparation

Download the dataset: Office31, OfficeHome, VisDA and DomainNet (real, painting and sketch).

Prepare dataset in data directory as follows

/path/to/your/dataset/images/amazon/      # Office
/path/to/your/dataset/RealWorld/          # OfficeHome 
/path/to/your/dataset/train/              # VisDA synthetic images
/path/to/your/dataset/test/               # VisDA real images
/path/to/your/dataset/sketch/             # DomainNet

For OfficeHome dataset, make sure that your folder name is RealWorld instead of Real World.
Modify root_path with /path/to/your/dataset/ in config files ./config/<dataset>-config.yaml.
Make a log directory by mkdir ./log.
Make a model directory by mkdir ./model. Download ImageNet pretrained model from Google Drive, then put the downloaded model into ./model.

Getting started

Train with command line (take office for example)

python main.py --gpu 0 --exp office31 --dataset office31 --source amazon --target dslr

Train with script

Modify ./config/<dataset>.sh:
- delete the lines which begin with #SBATCH
- specify $CUDA_VISIBLE_DEVICES
then
```
cd ./script
sh office31.sh          # or officehome/visda/domainnet
```
Train with Slurm script

Modify ./config/<dataset>.sh:
- #SBATCH [email protected]
- #SBATCH -p YOUR_partition
then
```
cd ./script
mkdir output
sbatch office31.sh      # or officehome/visda/domainnet
```
Monitor (TensorBoard required)
```
tensorboard --logdir=./log --port xxxx
```

Test with command line (take office for example)

python eval.py --gpu 0 --dataset office31 --source amazon --target dslr --model_path /path/to/your/model/final.pkl

Checkpoints

We provide the checkpoints for Office, OfficeHome, VisDA and DomainNet at Google Drive.

Citation

If you find this repository useful in your research, please consider citing:

@inproceedings{
chang2022unified,
title={Unified Optimal Transport Framework for Universal Domain Adaptation},
author={Wanxing Chang and Ye Shi and Hoang Duong Tuan and Jingya Wang},
booktitle={Advances in Neural Information Processing Systems},
editor={Alice H. Oh and Alekh Agarwal and Danielle Belgrave and Kyunghyun Cho},
year={2022},
url={https://openreview.net/forum?id=RTan64GlCLV}
}

uniot-for-unida's People

Contributors

Stargazers

Watchers

uniot-for-unida's Issues

Threading issues

When I run main.py, I get the following error:
RuntimeError:
An attempt has been made to start a new process before the
current process has finished its bootstrapping phase.

    This probably means that you are not using fork to start your
    child processes and you have forgotten to use the proper idiom
    in the main module:

        if __name__ == '__main__':
            freeze_support()
            ...

    The "freeze_support()" line can be omitted if the program
    is not going to be frozen to produce an executable.
w.start()