Pretrained model weights for the full pipeline described in the paper "Automated Multi-Class Structural Damage Detection and Segmentation for UAV-Based Infrastructure Inspection Using Deep Learning."
Files:
- best_resnet50_multiclass.pth — ResNet50 classifier, 4 damage classes (crack, corrosion, spalling, intact), 85.26% accuracy
- best_deeplabv3_4class.pth — DeepLabV3+ segmentation, 4 classes
- best_deeplabv3_segmentation.pth — DeepLabV3+ binary segmentation
Framework: PyTorch