CrackSwinT - Swin Transformer Based Crack Detection

This is the official repo for the paper - Swin Transformer Based Crack Detection (draft version). The paper is submitted to IEEE TITS and is in the review stage currently. In this paper, we leverage Shifted Window (Swin-T) architecture for crack segmentation using a hierarchical architecture. This way, we capture both global and local features of cracks

Datasets

We have used 3 datasets:-

1. CFD Dataset
2. CrackTree200
3. Crack500

Comparison with the current state-of-the-art

Here, we are comparing our results with CrackFormer-II

Note - Full code will be released once the paper is accepted.