WP-CrackNet: A Collaborative Adversarial Learning Framework for End-to-End Weakly-Supervised Road Crack Detection

TL;DR

WP-CrackNet introduces an end-to-end weakly-supervised framework for pixel-wise road crack detection using only image-level labels. It jointly optimizes a classifier, a reconstructor, and a detector in an adversarial mutual-learning loop, augmented by a path-aware attention module (PAAM) and a center-enhanced CAM consistency module (CECCM) to produce robust crack CAMs and reliable pseudo-labels. Empirical results on three datasets show detection performance close to supervised methods and superior to existing weakly-supervised approaches, with improved CAM quality and pseudo-label reliability due to online updating and CRF refinement. This approach significantly improves scalability of road inspection by reducing annotation costs while delivering accurate, pixel-level crack maps suitable for real-world maintenance decisions.

Abstract

Road crack detection is essential for intelligent infrastructure maintenance in smart cities. To reduce reliance on costly pixel-level annotations, we propose WP-CrackNet, an end-to-end weakly-supervised method that trains with only image-level labels for pixel-wise crack detection. WP-CrackNet integrates three components: a classifier generating class activation maps (CAMs), a reconstructor measuring feature inferability, and a detector producing pixel-wise road crack detection results. During training, the classifier and reconstructor alternate in adversarial learning to encourage crack CAMs to cover complete crack regions, while the detector learns from pseudo labels derived from post-processed crack CAMs. This mutual feedback among the three components improves learning stability and detection accuracy. To further boost detection performance, we design a path-aware attention module (PAAM) that fuses high-level semantics from the classifier with low-level structural cues from the reconstructor by modeling spatial and channel-wise dependencies. Additionally, a center-enhanced CAM consistency module (CECCM) is proposed to refine crack CAMs using center Gaussian weighting and consistency constraints, enabling better pseudo-label generation. We create three image-level datasets and extensive experiments show that WP-CrackNet achieves comparable results to supervised methods and outperforms existing weakly-supervised methods, significantly advancing scalable road inspection. The source code package and datasets are available at https://mias.group/WP-CrackNet/.

Figures (10)

• Figure 1: Road inspection for intelligent infrastructure maintenance with the proposed WP-CrackNet.
• Figure 2: The overall architecture of WP-CrackNet, consisting of a classifier, a reconstructor, and a detector. During the training phase, the classifier generates CAMs to localize discriminative regions, the reconstructor measures feature inferability, and the detector learns from pseudo labels derived from post-processed crack CAMs to produce pixel-wise road crack detection results.
• Figure 3: The proposed PAAM, consisting of a spatial attention branch and a channel attention branch.
• Figure 4: Ablation study results illustrating crack CAMs (b, d, f) obtained under different training strategies — classifier only, adversarial training, and adversarial training with CECCM — and their corresponding pseudo labels (c, e, g) produced via denseCRF, with (a) showing the input image.
• Figure 5: Examples of experimental results on the (I) CFD shi2016automatic, (II) DeepCrack liu2019deepcrack, and (III) Crack500 yang2019feature datasets: (a) Input images; (b) Ground Truth; (c) Deepcrack19 liu2019deepcrack; (d) Deepcrack18 zou2018deepcrack; (e) Crack-Att xu2023crack; (f) HrSegNet li2023real; (g) CT-crackseg tao2023convolutional; (h) AEFT yoon2022adversarial; (i) OC-CSE kweon2021unlocking; (j) ACR kweon2023weakly; (k) WS-SCS konig2022weakly; (l) UP-CrackNet ma2024up; (m) WP-CrackNet.