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
aspython tiff_read.py
. The masks are saved in the folderMasks
. -
Then, run
model.py
aspython 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.