Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CrossDiff: Diffusion Probabilistic Model With Cross-conditional Encoder-Decoder for Crack Segmentation

Xianglong Shi, Yunhan Jiang, Xiaoheng Jiang, Mingling Xu, Yang Liu

TL;DR

CrossDiff tackles the challenging problem of slender crack segmentation in industrial concrete by marrying diffusion probabilistic modeling with a cross-conditional encoder-decoder. The approach introduces a cross-shaped diffusion architecture that leverages a ViT-based Cross Encoder and a six-layer Cross Decoder, connected through a Fusion Nexus to condition the diffusion process on image priors while preserving fine crack details and semantic features. Empirical results across five crack datasets show sizable improvements (around 8 percentage points in IoU and Dice) over prior methods, with ablations confirming the contributions of both the Cross Encoder and Cross Decoder. This diffusion-based, cross-conditioned framework promises more robust and accurate crack delineation and offers a valuable direction for practical maintenance and automated inspection workflows.

Abstract

Crack Segmentation in industrial concrete surfaces is a challenging task because cracks usually exhibit intricate morphology with slender appearances. Traditional segmentation methods often struggle to accurately locate such cracks, leading to inefficiencies in maintenance and repair processes. In this paper, we propose a novel diffusion-based model with a cross-conditional encoder-decoder, named CrossDiff, which is the first to introduce the diffusion probabilistic model for the crack segmentation task. Specifically, CrossDiff integrates a cross-encoder and a cross-decoder into the diffusion model to constitute a cross-shaped diffusion model structure. The cross-encoder enhances the ability to retain crack details and the cross-decoder helps extract the semantic features of cracks. As a result, CrossDiff can better handle slender cracks. Extensive experiments were conducted on five challenging crack datasets including CFD, CrackTree200, DeepCrack, GAPs384, and Rissbilder. The results demonstrate that the proposed CrossDiff model achieves impressive performance, outperforming other state-of-the-art methods by 8.0% in terms of both Dice score and IoU. The code will be open-source soon.

CrossDiff: Diffusion Probabilistic Model With Cross-conditional Encoder-Decoder for Crack Segmentation

TL;DR

CrossDiff tackles the challenging problem of slender crack segmentation in industrial concrete by marrying diffusion probabilistic modeling with a cross-conditional encoder-decoder. The approach introduces a cross-shaped diffusion architecture that leverages a ViT-based Cross Encoder and a six-layer Cross Decoder, connected through a Fusion Nexus to condition the diffusion process on image priors while preserving fine crack details and semantic features. Empirical results across five crack datasets show sizable improvements (around 8 percentage points in IoU and Dice) over prior methods, with ablations confirming the contributions of both the Cross Encoder and Cross Decoder. This diffusion-based, cross-conditioned framework promises more robust and accurate crack delineation and offers a valuable direction for practical maintenance and automated inspection workflows.

Abstract

Crack Segmentation in industrial concrete surfaces is a challenging task because cracks usually exhibit intricate morphology with slender appearances. Traditional segmentation methods often struggle to accurately locate such cracks, leading to inefficiencies in maintenance and repair processes. In this paper, we propose a novel diffusion-based model with a cross-conditional encoder-decoder, named CrossDiff, which is the first to introduce the diffusion probabilistic model for the crack segmentation task. Specifically, CrossDiff integrates a cross-encoder and a cross-decoder into the diffusion model to constitute a cross-shaped diffusion model structure. The cross-encoder enhances the ability to retain crack details and the cross-decoder helps extract the semantic features of cracks. As a result, CrossDiff can better handle slender cracks. Extensive experiments were conducted on five challenging crack datasets including CFD, CrackTree200, DeepCrack, GAPs384, and Rissbilder. The results demonstrate that the proposed CrossDiff model achieves impressive performance, outperforming other state-of-the-art methods by 8.0% in terms of both Dice score and IoU. The code will be open-source soon.
Paper Structure (18 sections, 4 equations, 3 figures, 3 tables)

This paper contains 18 sections, 4 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Discriminant Segmentation
  4. Generative Segmentation
  5. Method
  6. Towards Generative Slender Segmentation
  7. Cross-Conditional Encoder-Decoder
  8. Cross Encoder
  9. Cross Decoder
  10. Fusion Nexus Module
  11. Optimization Objectives
  12. Experiments
  13. Dataset
  14. Evaluation Metrics
  15. Implementation Details
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Comparison between natural and conventional scene images with industrial crack images. The top row showcases images from mainstream semantic segmentation datasets, including Pascal VOC everingham2015pascal, MVTecAD bergmann2019mvtec, SegTrack-v2 li2013video and ISIC gutman2016skin, with corresponding ground truth images in the second row. The third row displays samples from CFD shi2016automatic, CrackTree-200 zou2012cracktree, DeepCrack liu2019deepcrack, GAPs384 eisenbach2017get, and Rissbilder pak2021crack with their ground truth images below.
  • Figure 2: An illustration of CrossDiff. CrossDiff consists of Diffusion backbone and cross-conditional encoder-decoder, including Cross Encoder, Fusion Nexus, and Cross Decoder, which jointly constitute the cross-shaped diffusion model structure.
  • Figure 3: The sequence of images, progressing from top to bottom, includes the input image, SegDecNet++ tabernik2023automated, CrossDiff, and ground truth mask (GT). It is apparent from the visual comparison that our model surpasses SegDecNet++ tabernik2023automated in terms of the accuracy of cracks localization.