Skip to content

Latest commit

 

History

History
25 lines (14 loc) · 913 Bytes

readme.md

File metadata and controls

25 lines (14 loc) · 913 Bytes

This is the submission by the team from IIT Madras which participated in Inter IIT Tech Meet 2018 conducted by IIT Bombay.

Problem statement

The goal was to achieve satellite image segmentation using a training dataset of only 14 images, of size approx 1200x1200.
The output image pixels had to be segmented into 8 classes ( or as background): Swimming Pool, Oceans, Grasslands, Forests, Roads, Railways, Bare soil and Buildings. More details are available in the Problem-Statement pdf titled "Eye-in-the-Sky"

Our Solution

We used a mixture of pure computer vision techniques and deep learning approaches for the problem. Our approach was to obtain binary segmentation maps class-wise. Details are mentioned in the report titled "InterIIT_2018IITM.pdf"

Code dependencies

  • python 3.5 or higher
  • Tensorflow 1.10 or higher
  • Keras
  • OpenCV 3.0.0 or higher
  • Numpy, Sklearn and matplotlib