Deep-Satellite

data

The images come from dstl 10 polygon annotated image in total

train_x and train_y

use vis.py to draw polygon in train_y and visualize the 3 bands images train_x

crop.py

stride = 100 crop 10 image and do 4 fold augumentation to 43560

drawSeg.py

a manual annotation tool left button down to draw a node right button down to fill the polygon
s to save
Esc to quit
remember to use F5 to refresh after drawing

make dataset

$ python tmake.py

then data.tfrecords

test dataset

read_tfrecorder.py to read and display the dataset

pretrain model

Download vgg19.mat

then move it to /Deep-Satelitte

train

$ python FCN.py

Acknowledgement

• FCN.tensorflow

• DeepLab-LargeFOV implemented in tensorflow