Satellite-Imagery-Feature-Detection

Introduction

• In this project, we aim to achieve segmentation of satellite image to detect various features in the image, as part of the DSTL challenge on Kaggle.

Compatibility

• This code has been tested on Ubuntu 16.04 LTS and mac-OSX.
• Dependencies - Python 3.5+, OpenCV 3.0+, Tensorflow.

Trained Checkpoint

• We have trained the model for a specific class of buildings,
• You can download the checkpoint here.

Usage

• Clone this repository by typing git clone https://github.com/atulapra/Satellite-Imagery-Feature-Detection.git in the terminal.

• Enter the cloned folder.

• Download the three band data from here and unzip it.

• Execute tiff_read.py as python tiff_read.py. The masks are saved in the folder Masks.

• Then, run model.py as python model.py.

• When you run it in train mode, training process takes place and checkpoint is saved in outputs folder.

• When you run it in test mode, the output images are saved in outputs folder.