What question did this study set out to answer?

This work aims to develop a hybrid architecture for efficient crack segmentation in images of civil infrastructure.

March 8, 2026SensorsOpen Access

A Novel CNN–ViT Model with Cascade Upsampling for Efficient Crack Segmentation

View Full Paper

Ask AI

Bookmark

Authors

Hybrid architecture improves crack segmentation in civil infrastructure imaging, indicating enhanced efficiency and lower latency.

This work aims to develop a hybrid architecture for efficient crack segmentation in images of civil infrastructure.
Introduced a CNN–ViT model combining a convolutional encoder and a Transformer bottleneck.
Utilized a progressive decoder with multi-level skip-feature fusion for fine spatial details restoration.
Trained with a composite Dice–Binary Cross-Entropy objective to address foreground sparsity.
Demonstrated improvements over traditional convolutional, Transformer-based, and hybrid segmentation methods.
Validated on four public benchmarks with consistent performance gains.
Showed favorable latency and memory usage suitable for real-time deployment.

Tibermacine et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark