The official implementation of the paper An Automated Method for the Creation of Oriented Bounding Boxes in Remote Sensing Ship Detection Datasets
Given images with objects with Horizontal Bounding Box annotations, the proposed method can create reliable Oriented Bounding Boxes, with a pipeline consisting of the Segment Anything Model (SAM), a morphological closing and a contour detection operation.
Additionally, this method can be used to create augmented versions of the dataset in a manner that resolves the objects' orientation imbalance.
Firstly, an anaconda environment needs to be set with a python>=3.9
conda create -n "env name" python="3.9 or above"
conda activate "env name"
Then install pytorch, torchvision and the segment anything model
conda install pytorch torchvision
pip install git+https://github.com/facebookresearch/segment-anything.git
Initially, clone the repository.
git clone https://github.com/GSavathrakis/hbb_to_obb.git
cd hbb_to_obb
Download the segment-anything model checkpoint from the official segment-anything model repo
To use the method for the creation of OBBs from HBBs , run
python OBB_generation/generate.py --dataset "dataset name" --image_path "The path to the images directory" --annotation_path "The path to the annotations directory" --sam_checkpoint_path "The path to where the segment-anything checkpoint is stored" --new_annotations_dir "The path where the newly created OBB annotations will be saved" --gen_mode
For the creation of augmented datasets with uniform object orientation distribution, run
python Augmentation/augm.py --image_path "The path to the images directory" --annotation_path "The path to the annotations directory" --aug_image_path "The path to the directory where the augmented images will be saved" --aug_annotation_path "The path to the directory where the annotations of the augmented images will be saved" --dataset_type "dataset name" --augm_method "The augmentation method to be used"
If you find this work useful for your research, please cite
@InProceedings{Savathrakis_2024_WACV,
author = {Savathrakis, Giorgos and Argyros, Antonis},
title = {An Automated Method for the Creation of Oriented Bounding Boxes in Remote Sensing Ship Detection Datasets},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops},
month = {January},
year = {2024},
pages = {830-839}
}