khoidoo/ibla

IBLA - Imbalance Learning Archive

IBLA - Imbalance Learning Archive

🔔🔔🔔 This project is shutdown owing to a very bad coding architecture, the new version of this project will be release soon.

Setup

Environment

Create a virtual environment and upgrade the version of pip

python3 -m venv .env
source .env/bin/activate
python -m pip install -U pip

Install the required packages

cd /path/to/cloned/repo/directory
pip install -r requirements.txt

Install Pytorch packages, the version of the Nvidia driver is 535 and the CUDA version is 12.1 on a Ubuntu-based system

pip3 install torch torchvision torchaudio

Dataset

Contribute to this project

Folder Structure

• dataset
  • [ds].py where ds is any dataset name
  • getds.py used as a mapping from args to a dataset
• loss
  • [loss].py where loss is any loss name
• metrics
  • [metric].py where metric is any metric name
• model
  • [model_name]
    • *.py
• utils
• main.py
• mapping.py
• requirements.txt
• trainer.py

Dataset Contribution

• Data sources and related annotations must be put into dataset/source folder, which is an ignored folder. A custom dataset class inheriting torch.utils.data.Dataset of Pytorch and read all data from dataset/source.
• ⚠️⚠️: Do not change the content of any file inside dataset/, just create a new .py file to contribute your custom dataset.

Model/Loss Contribution

• Create a new folder/new file .py to store your model architecture/loss function and its sub-components
• ⚠️⚠️: Do not change the content of any file inside model/ and loss/

Workflow

• For collaborators, create a new branch with the name as follows: <task>-<name> (i.e. model_deeplabv3) and then pull a request to merge to branch main.
• For outer collaborators, fork this repo, and then pull a request also.
• For review the pull request, you can tag the owner of this repo, the owner will merge and complete the final task to emerge your contribution thoroughly.

Citation

ibla's Issues

I have a problem displaying img and msk.

In CustomBagls, init function, I don't use self.aug_transforms, but img and msk don't match. When Custom follows the dataset Drive, it outputs correctly.