Granite is a lightweight, pre-trained MobileNet model for detected cracks on concrete walls, slabs, pavements and other stuctures made of concrete. This pre-trained model was created using transfer learning and has a validation accuracy of 99.95 %. The model was trained on the Concrete Crack Images for Classification from Medeley.

Scripts written and trained by Nikhil Raghavendra (@nikhilraghava).

• This repository
• Python 3.6.5
• TensorFlow 1.9.0
• NumPy 1.14.2

Clone or download this repository and execute the following command in the terminal with this repository being the current working directory.

python classify.py [path to image — relative to the script]

• macOS 10.13.6
• macOS 10.14

Licensed under MIT License.