Official repository for: CrackMamba: Topology-aware Mamba for Crack Segmentation in Structures
Step-1: Create a new conda environment & install requirements
conda create -n crack_mamba python=3.10
conda activate crack_mamba
pip install torch==2.0.1 torchvision==0.15.2
pip install causal-conv1d==1.1.1
pip install mamba-ssm
pip install torchinfo timm numbaStep-2: Install CrackMamba
git clone https://github.com/shengyu27/CrackMamba
cd CrackMamba
pip install -e .Dataset:
CrackSeg9K:
We use the same data & processing strategy following SwinUMamba. We downloaded version 2.0 of the dataset and re-screened it, see the file in the folder for the exact dataset split:
/CrackMamba/dataset/train.txt
/CrackMamba/dataset/test.txt
Download dataset from here and put them into the data folder. Then preprocess the dataset with following command:
nnUNetv2_plan_and_preprocess -d DATASET_ID --verify_dataset_integritySewerCrack:
The dataset was derived from a series of original CCTV videos provided by a municipal authority in the southern United States. The process involved extracting frames from these videos and manually annotating them. Due to privacy concerns and copyright restrictions, we regret that we are unable to provide a public access link to this dataset.
CHASE_DB1:
For your convenience we have provided the data for this dataset in the dataset folder for you to download.
It's worth noting that the pixel values in mask are only 0 and 1, so they look all black when visualized. If any subsequent manipulation is required, you can handle it yourself.
Pretrained model:
You can download the model weights here.
Using the following command to train & evaluate CrackMamba
bash scripts/train_Crack.shHere, You can configure the content of the script, such as folder address, dataset number, etc., to train on different datasets.
@article{ZUO2024105845,
title = {Topology-aware mamba for crack segmentation in structures},
journal = {Automation in Construction},
volume = {168},
pages = {105845},
year = {2024},
issn = {0926-5805},
doi = {https://doi.org/10.1016/j.autcon.2024.105845},
url = {https://www.sciencedirect.com/science/article/pii/S0926580524005818},
author = {Xin Zuo and Yu Sheng and Jifeng Shen and Yongwei Shan},
keywords = {Crack segmentation, Mamba, Snake scan, CrackSeg9k, SewerCrack, CHASE_DB1},
abstract = {CrackMamba, a Mamba-based model, is designed for efficient and accurate crack segmentation for monitoring the structural health of infrastructure. Traditional Convolutional Neural Network (CNN) models struggle with limited receptive fields, and while Vision Transformers (ViT) improve segmentation accuracy, they are computationally intensive. CrackMamba addresses these challenges by utilizing the VMambaV2 with pre-trained ImageNet-1 k weights as the encoder and a newly designed decoder for better performance. To handle the random and complex nature of crack development, a Snake Scan module is proposed to reshape crack feature sequences, enhancing feature extraction. Additionally, the three-branch Snake Conv VSS (SCVSS) block is proposed to target cracks more effectively. Experiments show that CrackMamba achieves state-of-the-art (SOTA) performance on the CrackSeg9k and SewerCrack datasets, and demonstrates competitive performance on the retinal vessel segmentation dataset CHASE_DB1, highlighting its generalization capability.}
}We thank the authors of nnU-Net, SwinUMamba, Mamba and VMamba for making their valuable code & data publicly available.