Road-Crack-Segmentation--Keras

The project uses Unet-based improved networks to study road crack segmentation, which is based on keras.
GCUnet is an improvement to Unet that introduces the Global Context Block in Unet.

Requirements

• python 3.6.8
• tensorflow-gpu 1.8
• Keras 2.2.4
• opencv
• tqdm
• numpy
• glob
• argparse
• matplotlib

Usage

1. Download dataset

CRACK500, key: j54r

2. Train

eg. python train.py --train_images dataset/CRACK500/traincrop/ --train_annotations dataset/CRACK500/traincrop/ --epoch 100 --batch_size 32

3. Test

eg. python test.py --save_weights_path 'checkpoint/'+ 'Unet/' + 'weights-099-0.1416-0.9787.h5 --vis False

Results