Skip to content

This is the winning code of sattelite image segmentation challenge Eye-In-The-Sky

Notifications You must be signed in to change notification settings

CaptainDredge/Eye-In-The-Sky-Image_Segmentation_Challenge

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Eye In The Sky

Winning code of Eye In The Sky Satellite image segmentation Competition as a part of Inter-IIT tech meet hosted by IIT Bombay.

Prerequisites

torch
tensorboardX
tensorflow
scikit-image
scikit-learn
tifffile
skimage
numpy 
argparse
gflags

Installing

Basic dependencies stated in requirements.txt

To install dependencies: pip install -r requirements.txt

File Description

1.aug.py - applies augmentation to the training data and saves it in a folder named image

python aug.py --path=[train folder] --dest=[dest folder]
  1. Patches.py creates patches of the given shape and saves them to a specified location
python patches.py --source=[source folder] --dest=[destiantion folder] --height=256 --width=256 --stride=0.5 --mode=train
  1. train.py - trains model
python train.py --id=[experiment id] --epochs=25 --resume=[restart training(true/false)] --batch_size=16 --lr=0.001 --tag=[tag for tensorboard runs] --gpu=[true/false]

How to run with default parameters

A typical top-level directory layout

.
├── train                   # Contains original sat and gt files 
├── image                    # contains images after augmentation
├── data                     # contains patches 
.
.         
└── README.md

copy train folder as given and should have sat/ and gt/ folder

Then run the following process

process:

python aug.py
python channel9.py
python patches.py
python train.py --id=[] --tag=[]
python predict.py --id=[] --sub_id=[]
python stitch.py --id=[] --su_id=[]

For repeating the predictions(Final Model)

  1. Copy test folder as given
2. Run python patches.py --mode='test'
3. Then python predict.py --id=1 --sub_id=1
4. Then python stitch.py --id=1 --sub_id=1

Final Predictions will be saved in TestStichedid_1

Instrucions for binary Models (Not Our Final Model)

Note - Copy given dataset folder as The-Eye-in-the-sky-dataset, The-Eye-in-the-sky-test-dataset in the Binary folder

1. python patches.py
2. python create_binary_mask.py
3. python train.py --id=[] 
4. python predict.py --id=[] --sub_id=[]
5. python stich_mask.py --id=[] --su_id=[]

About

This is the winning code of sattelite image segmentation challenge Eye-In-The-Sky

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published