Image-segmentation

Binary Class Image Segmentation

This project focuses on binary class image segmentation of buildings from satellite-captured aerial imagery. The goal is to classify buildings in the imagery, separating them from the background.

Dataset

• The dataset consists of 137 images and 137 corresponding masks for training.
• Additionally, there are 10 test images and 10 test masks for evaluation.
• All images are captured via satellite and consist of aerial views of buildings, segmented into two classes: buildings and non-buildings.

Project Aim

The primary objective is to implement binary class image segmentation to correctly identify and classify buildings in the satellite images. This helps in extracting accurate boundaries and structures of buildings from the aerial imagery.

Output

Multiclass Image Segmentation of Aerial Imagery

This project involves multiclass image segmentation of aerial imagery captured by satellites. The dataset includes pixel-wise semantic segmentation into 6 different classes.

Dataset

• Total Images: 72 images, grouped into 6 larger tiles.
• Classes: The images are segmented into 6 distinct classes, each represented with a unique color code to facilitate differentiation.

Project Aim

The primary objective is to classify all the different classes in the aerial imagery using multiclass image segmentation techniques.

Preprocessing Steps

1. Combine Image and Mask Directories:

   • All images and masks from each individual tile folder were combined to create separate directories for images and masks.

2. Data Augmentation:

   • Augmentation techniques were applied to increase the number of input images in the dataset, enhancing the model's ability to generalize.

3. Train-Test Split:

   • The preprocessed images were split into training and testing sets to evaluate the model's performance.

4. One-Hot Encoding:

   • One-hot encoding was performed on the images to prepare them for multiclass classification.

5. Data Generators:

   • Data generators were created to facilitate training and validation processes. These generators run in real-time, forming batches of images and applying one-hot encoding to them.

Notes

• The color-coded classes help in distinguishing between different segments of the aerial images, improving the clarity and accuracy of the segmentation task.

Output

How to Run Streamlit

streamlit run main.py