Deep Augmented Neural Network for Pavement Crack Segmentation
This repository contains trained model reported in the paper:
V.Polovnikov, D. Alekseev, I. Vinogradov, G. Lashkia, DAUNet: Deep Augmented Neural Network for Pavement Crack Segmentation, IEEE Access, Vol.9, 2021. [https://ieeexplore.ieee.org/document/9531629]
git clone https://github.com/dvalex/daunet
cd daunet/python
pip install -r requirements.txt
Unix users can use data/download.sh script to automate:
cd daunet/data
bash download.sh
For training: download crack500.zip from Google Drive Unzip it into data/cracks500 subfolder
For evaluating: download testcrop.zip from Google Drive Unzip it into data/testcrop subfolder
cd daunet/python
export SM_FRAMEWORK=tf.keras
python train.py
python finetune.py
To run inference at all images in directory (by default data/testcrop) run
cd daunet/python
python inference.py
After that one can calculate AIU, ODS, OIS, sODS, sOIS using matlab evaluation scripts