The PyTorch implementation of the skea-topo aware loss proposed in paper: Enhancing Boundary Segmentation for Topological Accuracy with Skeleton-based Methods.
This paper have been accepted by IJCAI2024.
Topological consistency plays a crucial role in the task of boundary segmentation for reticular images, such as cell membrane segmentation in neuron electron microscopic images, grain boundary segmentation in material microscopic images and road segmentation in aerial images. In these fields, topological changes in segmentation results have a serious impact on the downstream tasks, which can even exceed the misalignment of the boundary itself. To enhance the topology accuracy in segmentation results, we propose the Skea-Topo Aware loss, which is a novel loss function that takes into account the shape of each object and topological significance of the pixels. It consists of two components. First, the skeleton-aware weighted loss improves the segmentation accuracy by better modeling the object geometry with skeletons. Second, a boundary rectified term effectively identifies and emphasizes topological critical pixels in the prediction errors using both foreground and background skeletons in the ground truth and predictions. Experiments prove that our method improves topological consistency by up to 7 points in VI compared to 13 state-of-art methods, based on objective and subjective assessments across three different boundary segmentation datasets.
python 3.8
pytorch 1.13.0+cu116
gala (for evaluation)
gala is installed according to https://github.com/janelia-flyem/gala.
- SNEMI3D can be download from https://zenodo.org/record/7142003, and it should be placed in ./data/snemi3d/.
- IRON can be download from https://github.com/Keep-Passion/pure_iron_grain_data_sets, and it should be placed in ./data/iron/.
- MASS ROAD can be download from https://www.kaggle.com/datasets/balraj98/massachusetts-roads-dataset, and it should be placed in ./data/mass_road/.
Usage Demo:
# prepare datasets and calculate mean and std for each datasets
python ./data/data_generator.py
# generate skeleton aware weighted map:
python ./data/skeleton_aware_loss_gen.py
# generate foreground and background skeletons for labels
python ./data/skeleton_gen.py
# train
run train.ipynb
# eval
run eval.ipynb
The example results are shown as follows:
If you find our work is useful in your research or applications, please consider giving us a star 🌟 and citing it.
@misc{liu2024enhancing,
title={Enhancing Boundary Segmentation for Topological Accuracy with Skeleton-based Methods},
author={Chuni Liu and Boyuan Ma and Xiaojuan Ban and Yujie Xie and Hao Wang and Weihua Xue and Jingchao Ma and Ke Xu},
year={2024},
eprint={2404.18539},
archivePrefix={arXiv},
primaryClass={cs.CV}
}