Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
May 3, 2026

LiteCrackSeg: A lightweight hybrid CNN-transformer for efficient crack segmentation.

View Full Paper
Ask AI
Bookmark
Share

Authors

KGKaleb Amsalu GobenaCentral South UniversityMRM D Youshuf Khan RakibCentral South UniversityFTFiseha Berhanu TesemaUniversity of Nottingham Ningbo China

Discussion

Loading...

Member takes

Implication

Randomized trial evaluates a hybrid model for crack segmentation in infrastructure, suggesting improved accuracy and efficiency for edge deployment.

Key Points

  • This research aims to develop a lightweight hybrid CNN-transformer model for efficient crack segmentation in structural health monitoring.
  • Introduced LiteCrackSeg, a hybrid MobileViT architecture with a Morphology-Aware bottleneck.
  • Utilized Dynamic Snake Convolutions to enhance sensitivity to crack shapes.
  • Implemented Tversky loss to address class imbalance during training.
  • Achieved state-of-the-art segmentation performance across three datasets.
  • Required only 2.72M parameters and 3.23 GFLOPs for real-time inference.
  • Reached 56 FPS on 512 × 512 images for effective edge device deployment.

Cite This Study

Gobena et al. (2026) studied this question.

synapsesocial.com/papers/69f6e6648071d4f1bdfc71b2https://doi.org/10.1371/journal.pone.0347765
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1A Lightweight CNN–Mamba Hybrid Architecture for Efficient Crack Segmentation2026
  2. 2A Novel CNN–ViT Model with Cascade Upsampling for Efficient Crack Segmentation2026 · 2 citations
  3. 3Enhanced Crack Segmentation via Dual-Branch CNN-Transformer Architecture with Linear Perception and Multi-Scale Refinement2025
  4. 4Enabling Real-Time, Cost-Efficient, and Lightweight High-Speed Crack Segmentation using Self-Supervised Attention Mechanism2026
  5. 5CSegNet：A Crack Segmentation Network Combining CNN and Transformer2024 · 5 citations