 
  Create new CONDA environment and install necessary packages
    conda create -n l6gd python=3.9
    conda activate l6gd
    conda install pip
    pip install -r requirements.txt
You can request for our HuggingFace dataset at our project page.
To start training the model, run
    python3 train.py --config <path to the config file>
Config files are stored in ./config. Remember to change dataset_path in the config files after downloading the dataset. After training, log files and model weights will be saved to ./log.
To detect grasp for test data, run
    python3 generate.py --config <path to the config file> --checkpoint <path to the trained model> --data_path <path to the downloaded dataset> --n_sample 64
The detected grasp poses will be saved to an all_data.pkl file in the corresponding log directory.
For evaluation, excecute
    python eval.py --data <path to all_data.pkl>
where <path to all_data.pkl> is the path to the file all_data.pkl generated after the step of detecting grasps.
If you find our work interesting or helpful for your research, please consider citing our paper as
    @inproceedings{nguyen2024language,
        title={Language-driven 6-dof grasp detection using negative prompt guidance},
        author={Nguyen, Toan and Vu, Minh Nhat and Huang, Baoru and Vuong, An and Vuong, Quan and Le, Ngan and Vo, Thieu and Nguyen, Anh},
        booktitle={ECCV},
        year={2024}
    }