DAUNet

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]

INSTALLATION

git clone https://github.com/dvalex/daunet
cd daunet/python
pip install -r requirements.txt

DOWNLOAD DATASET & DATA PREPARATION

Unix users can use data/download.sh script to automate:

cd daunet/data
bash download.sh

Manual

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

TRAINING

cd daunet/python
export SM_FRAMEWORK=tf.keras

First stage

python train.py

Second stage

python finetune.py

INFERENCE AND EVALUATION

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