Semantic Segmentation of cracks

This project employs CNN and transformer models for semantic segmentation of building damage in Mexico City's 2017 earthquake, using annotated imagery to identify features like cracks and exposed rebar.

Directory and File Descriptions

• README.md: Main documentation of the project
• requirements.txt: List of Python dependencies
• .gitignore: Git ignore file
• Damage dataset/: Images of structural damage and README
• Data Processing/: Data preprocessing scripts and notebooks
• CNN Models/: Convolutional neural network models that we used
• Transformer Models/: Transformer-based models that we used
• Project_Presentation.pptx: Contains presentation slides that offer a comprehensive overview of the project, including its aims, methods, and key results.

Data Sources

Images are gathered from:

1. DataCenterHub
2. Photographed by Vedhus Hoskere

Annotations

Two sets of annotations are available in PNG format for each image:

1. Fine damage and damage-like features, including features like cracks, exposed rebar, cables, etc Class names: 'No', 'Scratches', 'Grooves/Joints', 'Cables', 'Filled Cracks', 'Cracks', 'Exposed Rebar'

Class	Color
No	black
Scratches	red
Grooves/Joints	green
Cables	white
Filled Cracks	lightgrey
Cracks	red
Exposed Rebar	orange

2. Coarse damage and damage-like features, including spalling, dirt, etc. Class names: 'No', 'Shadows', 'Dirt', 'Vegetative Growth', 'Debris', 'Marks', 'Spalling', 'Voids'

Class	Color
No	black
Shadows	grey
Dirt	goldenrod
Vegetative Growth	springgreen
Debris	fuchsia
Marks	purple
Spalling	tomato
Voids	yellow

Use PIL to open images with single integer at each pixel as opposed to a color image

The index can be used to identify the corresponding class type

import numpy as np
import matplotlib.pyplot as plt
from PIL import Image

im = Image.open(os.path.join(folder,file))
image = np.array(im)
plt.subplot(image)