Introduction

• This repository contains a UNet implementation in tensorflow to segment out features from Satellite Images.
• The data was provided by DSTL as a part of its challenge in Kaggle.

Dependencies

• This code requires Python 2.7/3.5 , tensorflow (1.0+), OpenCV 3.0+
• This code has been tested on Ubuntu 16.04 LTS, macOS 10.11, Windows 10

Pretrained Checkpoint:

• I have trained the UNet for the specific case of buildings.
• You can download the checkpoint at: Checkpoint
• You can train the model for any class you require by changing the class number from 1 to whatever number you desire.

For more information on the dataset click Here