dinoDet

TL;DR: Combining DINO with a Region Proposal Network (RPN) for instance segmentation

We provide a minimal example of dinoDet for instance segmentation on a toy dataset. Our tutorial is based on the detectron2 Colab notebook.

Getting started

• We recommend using docker:

docker pull pytorch/pytorch:2.2.0-cuda12.1-cudnn8-devel
docker run -it --name detectron2 --shm-size=8gb --gpus all -e DISPLAY=$DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix  pytorch/pytorch:2.2.0-cuda12.1-cudnn8-devel bash
# inside the docker container:
apt update && apt install git wget ffmpeg libsm6 libxext6 unzip

• Install dependencies:

git clone https://github.com/facebookresearch/detectron2.git
cd detectron2 && git checkout a1ce2f956a1d2212ad672e3c47d53405c2fe4312 && cd ..
python -m pip install -e detectron2
pip install numpy==1.26.4
pip install opencv-python==4.11.0.86
pip install xformers==0.0.24

• Clone repository and download dataset:

git clone https://github.com/hsp-iit/dinoDet.git
cd dinoDet
wget https://github.com/matterport/Mask_RCNN/releases/download/v2.1/balloon_dataset.zip
unzip balloon_dataset.zip
rm balloon_dataset.zip

• Launch training:

cd dinoDet
PYTHONPATH=. python tools/lazyconfig_train_net.py --config-file configs/run.py

Citation

If you find our work useful, please consider citing our paper as follows:

@inproceedings{vasile2025continuous,
  title={Continuous Wrist Control on the Hannes Prosthesis: a Vision-based Shared Autonomy Framework},
  author={Vasile, Federico and Maiettini, Elisa and Pasquale, Giulia and Boccardo, Nicol{\`o} and Natale, Lorenzo},
  booktitle={2025 IEEE International Conference on Robotics and Automation (ICRA)},
  pages={},
  year={2025},
}

Mantainer

This repository is mantained by:

	@FedericoVasile1