architrk2112/Mamba

# GL-Mamba for Brain Tumor Segmentation This project implements a Mamba-based neural network architecture, named GL-Mamba, for medical image segmentation, specifically targeting the BraTS 2021 dataset for brain tumor segmentation. ## Project Structure - `glmamba/`: Contains the core source code for the project. - `data/`: Data loading and preprocessing utilities. - `losses/`: Implementation of loss functions. - `metrics/`: Evaluation metrics. - `models/`: The GL-Mamba model implementation. - `utils/`: Helper scripts for argument parsing, checkpoints, etc. - `train.py`: The main script for training the model. - `eval.py`: The script for evaluating a trained model. - `infer.py`: The script for running inference on new data. - `scripts/`: Contains example SLURM scripts for running training and evaluation on a cluster. - `requirements.txt`: A list of Python dependencies for this project. ## Installation 1. **Clone the repository:** ```bash git clone cd Mamba-LBP ``` 2. **Install dependencies:** It is recommended to use a virtual environment. ```bash python -m venv venv source venv/bin/activate pip install -r requirements.txt ``` ## Dataset This model is designed to work with the BraTS 2021 dataset. You will need to organize your data and create a JSON file that specifies the file paths for your training, validation, and testing sets. The training script expects the path to the data root directory and this JSON file. ## Usage The primary scripts for interacting with the model are `train.py`, `eval.py`, and `infer.py`. You can see all available command-line arguments in `glmamba/utils/argparse.py`. ### Training Here is an example command to start a training run. You will need to adjust the paths and hyperparameters according to your setup. ```bash python train.py \ --data_root_dir /path/to/brats2021/dataset/ \ --data_list_file_path /path/to/your/datalist.json \ --log_dir ./logs \ --batch_size 2 \ --num_workers 4 \ --learning_rate 1e-4 \ --d_model 192 ``` …