RUC-Net: A Residual-Unet-Based Convolutional Neural Network for Pixel-Level Pavement Crack Segmentation

This is my own PyTorch implementation of 2023 U-Net (RUC-Net) for crack segmentation. Learn more: https://www.mdpi.com/1424-8220/23/1/53

Architecture of RUC-Net

Usage

Just download RUC-Net.py + scse.py and import RUCNet model into your model initialization pipeline.

pytorch == 2.0.0

Experiments

Bias and limitations

If you would look through my code, you will notice 2 differents from original paperwork:

• Last conv layer were changed from conv3x3 to conv1x1. I suppose, it's a paperwork typo.
• GroupNorm instead of BatchNorm. It's preferable to use GN if your batch_size < 32, but if you don't need this, just change it in code.