udl-vis 0.3.2

unified pytorch framework for vision task

UDL

UDL is a unified pytorch framework for vision research:

• UDL has faster library loading speed and more convenient reflection mechanism to call different models and methods.
• UDL is based on MMCV which provides the following functionalities.
• UDL is based on NNI to peform automatic machine learning.

English | 简体中文

See the repo for more detailed descriptions.

Features

Requirements

• Python3.7+, Pytorch>=1.6.0
• NVIDIA GPU + CUDA
• Run python setup.py develop

Note: Our project is based on MMCV, but you needn't to install it currently.

Quick Start

Step0. We use UDL in PanCollection, first please set your Python environment.

git clone https://github.com/XiaoXiao-Woo/UDL

git clone https://github.com/XiaoXiao-Woo/PanCollection

Then,

python setup.py develop

or

pip install -i udl-vis https://pypi.org/simple

Step1.

• Download datasets (WorldView-3, QuickBird, GaoFen2, WorldView2) from the homepage. Put it with the following format.

• Verify the dataset path in PanCollection/UDL/Basis/option.py, or you can print the output of run_pansharpening.py, then set cfg.data_dir to your dataset path.

|-$ROOT/Datasets
├── pansharpening
│   ├── training_data
│   │   ├── train_wv3.h5
│   │   ├── ...
│   ├── validation_data
│   │   │   ├── valid_wv3.h5
│   │   │   ├── ...
│   ├── test_data
│   │   ├── WV3
│   │   │   ├── test_wv3_multiExm.h5
│   │   │   ├── ...

Step2. Open PanCollection/UDL/pansharpening, run the following code:

python run_pansharpening.py

step3. How to train/validate the code.

• A training example：

  run_pansharpening.py

  where arch='BDPN', and configs/option_bdpn.py has:

  cfg.eval = False,

  cfg.workflow = [('train', 50), ('val', 1)], cfg.dataset = {'train': 'wv3', 'val': 'wv3_multiExm.h5'}

• A test example:

  run_test_pansharpening.py

  cfg.eval = True or cfg.workflow = [('val', 1)]

Step4. How to customize the code.

One model is divided into three parts:

1. Record hyperparameter configurations in folder of PanCollection/UDL/pansharpening/configs/option_<modelName>.py. For example, you can load pretrained model by setting model_path = "your_model_path" or cfg.resume_from = "your_model_path".

2. Set model, loss, optimizer, scheduler in folder of PanCollection/UDL/pansharpening/models/<modelName>_main.py.

3. Write a new model in folder of PanCollection/UDL/pansharpening/models/<modelName>/model_<modelName>.py.

Note that when you add a new model into PanCollection, you need to update PanCollection/UDL/pansharpening/models/__init__.py and add option_.py.

Others

• if you want to add customized datasets, you need to update:

PanCollection/UDL/AutoDL/__init__.py.
PanCollection/UDL/pansharpening/common/psdata.py.

• if you want to add customized tasks, you need to update:

1.Put model_<newModelName> and <newModelName>_main in PanCollection/UDL/<taskName>/models.
2.Create a new folder of PanCollection/UDL/<taskName>/configs to put option_<newModelName>.
3.Update PanCollection/UDL/AutoDL/__init__.
4.Add a class in PanCollection/UDL/Basis/python_sub_class.py, like this:
class PanSharpeningModel(ModelDispatcher, name='pansharpening'):

• if you want to add customized training settings, such as saving model, recording logs, and so on. you need to update:

PanCollection/UDL/mmcv/mmcv/runner/hooks

Note that: Don't put model/dataset/task-related files into the folder of AutoDL.

• if you want to know more details of runner about how to train/test in PanCollection/UDL/AutoDL/trainer.py, please see PanCollection/UDL/mmcv/mmcv/runner/epoch_based_runner.py

Contribution

We appreciate all contributions to improving PanCollection. Looking forward to your contribution to PanCollection.

Citation

Please cite this project if you use datasets or the toolbox in your research.

@misc{PanCollection,
    author = {Xiao Wu, Liang-Jian Deng and Ran Ran},
    title = {"PanCollection" for Remote Sensing Pansharpening},
    url = {https://github.com/XiaoXiao-Woo/PanCollection/},
    year = {2022},
}

@InProceedings{Wu_2021_ICCV,
    author    = {Wu, Xiao and Huang, Ting-Zhu and Deng, Liang-Jian and Zhang, Tian-Jing},
    title     = {Dynamic Cross Feature Fusion for Remote Sensing Pansharpening},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2021},
    pages     = {14687-14696}
}

Acknowledgement

• MMCV: OpenMMLab foundational library for computer vision.

License & Copyright

This project is open sourced under GNU General Public License v3.0.