def get_images(self, imageId, img_key=None):
'''
Load images correspoding to imageId
Parameters
----------
imageId : str
imageId as used in grid_size.csv
img_key : str
{None, '3', 'A', 'M', 'P'}, optional
Specify this to load single image
None loads all images (i.e. every band of the same image) and returns in a dict
'3' loads image from three_band/
'A' loads '_A' image from sixteen_band/
'M' loads '_M' image from sixteen_band/
'P' loads '_P' image from sixteen_band/
Returns
-------
images : dict
A dict of image data from TIFF files as numpy array
'''
creator = gdal_utils()
img_names = self.get_image_names(imageId)
images = dict()
if img_key is None:
for k in img_names.keys():
images[k] = creator.gdal_to_nparr(img_names[k])
else:
images[img_key] = creator.gdal_to_nparr(img_names[img_key])
return images
def main():
''' Performs Pansharpening with M bands and Panchromatic Band.'''
# Base Address
inDir = os.getcwd()
# Train-images multiploygon co-ordinates
df = pd.read_csv(os.path.join(inDir, 'Data/train_wkt_v4.csv'))
print(df.head())
# Distinct imageIds in the DataFrame
trainImageIds = []
for i in range(len(df)):
string = (df.iloc[i, 0])
if string not in trainImageIds:
trainImageIds.append(string)
print("The are " + str(len(trainImageIds)) + " distinct ids." +
str(df.head()))
trainImageIds.sort()
creator = gdal_utils()
for imageID in trainImageIds:
string_base = "gdal_sharpen.py "
bands = "-b 1 -b 2 -b 3 -b 4 -b 5 -b 6 -b 7 -b 8 "
panchro = os.path.join(inDir,
"Data/sixteen_band/" + imageID + "_P.tif ")
print(panchro)
m_band = os.path.join(inDir,
"Data/sixteen_band/" + imageID + "_M.tif ")
print(m_band)
output = os.path.join(
inDir, "Data/pan_sharpened_images/" + imageID + "_M.tif")
print(output)
final = string_base + bands + panchro + m_band + output
print(final)
creator.gdal_pansharpen(final)
def generate_all_masks(self):
# Base Address
inDir = os.getcwd()
# Helper class
creator = gdal_utils()
# train-images multiploygon co-ordinates
df = pd.read_csv(os.path.join(inDir, 'Data/train_wkt_v4.csv'))
print(df.head())
# grid size will also be needed later..
gs = pd.read_csv(os.path.join(inDir, 'Data/grid_sizes.csv'),
names=['ImageId', 'Xmax', 'Ymin'],
skiprows=1)
print(gs.head())
# Distinct imageIds in the DataFrame
trainImageIds = self.get_distinct_ids(df)
for key, classes in enumerate(self.CLASSES):
base = os.path.join(os.getcwd(), 'Data_masks')
path = os.path.join(base, self.CLASSES[classes])
print(base)
print(path)
print(classes)
for imageId in trainImageIds:
image = self.get_images(imageId, 'P')
print("imageId : {}, image_shape : {}".format(
imageId, image['P'].shape))
mask = self.generate_mask_for_image_and_class(
image['P'].shape, imageId, classes, gs, df)
print("mask : {}".format(mask.shape))
ref_raster_fn = os.path.join(
os.getcwd(), "Data/sixteen_band/" + imageId + "_P.tif")
new_raster_fn = os.path.join(
path, imageId + "_" + self.CLASSES[classes] + ".tif")
temp = creator.create_tiff_file_from_array(
ref_raster_fn, new_raster_fn, mask * 255)
temp = None
def get_training_image_pair(files_train, imageId):
''' Gets the input,truth for an image.
Description :
Get image-ground-truth pair of the image with id "imageId" from the list of
training images "files_train"
Arguments :
files_train -- List.
The list of names of the tiff images that will be used for training
imageId -- String.
The id of the image.
ex. '6010_0_0'
Returns :
(image_train,truth) -- Tuple.
This tuple containing the input image and the ground truth.
'''
if imageId not in files_train:
raise ValueError("Invalid value of imageId")
return None
# Using gdal to read the input image
reader = gdal_utils()
path = os.path.join(os.getcwd(), "Data/image_stacks/" + imageId + ".tif")
image_train = reader.gdal_to_nparr(path)
if image_train is None:
print("Failed to load image. Wrong path name provided.")
return None
# Using gdal to read the ground truth
path = os.path.join(os.getcwd(),
"Data_masks/Fast_H20/" + imageId + "_Fast_H20.tif")
truth = reader.gdal_to_nparr(path)
if truth is None:
print("Failed to load groung truth. Wrong path name provided.")
return None
return (image_train, truth)
return thresh*255
############################################# MODEL BUILDING #############################################
if __name__ == '__main__':
files_train,files_test = get_masks_list()
(image_test,truth1) = get_testing_image_pair(files_test,files_test[1])
# Predictions and Results before Post Processing
answer = model(image_test,truth1,img_rows=112,img_cols=112,num_channels=9)
result = compute_jaccard(truth1/255,answer/255)
answer = answer.astype(dtype = np.uint8)
writer_1 = gdal_utils()
writer_1.create_tiff_file_from_array(os.path.join(os.getcwd(),"Data_masks/Road/Test/" + files_test[1] + "_Road.tif"),os.path.join(os.getcwd(),"Results/pred_Roads.tif"),answer)
# Post Processed
re = post_processing(answer/255)
result = compute_jaccard(truth1/255,re)
print("Post processing : {}".format(result))
re = re*255
re = re.astype(dtype = np.uint8)
plt.imshow(re[:,:,0])
writer_2 = gdal_utils()
writer_2.create_tiff_file_from_array(os.path.join(os.getcwd(),"Data_masks/Road/Test/" + files_test[1] + "_Road.tif"),os.path.join(os.getcwd(),"Results/processed_pred_Roads.tif"),re)
if __name__ == '__main__':
import pandas as pd
# Base Address
inDir = os.getcwd()
# train-images multiploygon co-ordinates
df = pd.read_csv(os.path.join(inDir, 'Data/train_wkt_v4.csv'))
print(df.head())
# Distinct imageIds in the DataFrame
trainImageIds = []
for i in range(len(df)):
string = (df.iloc[i, 0])
if string not in trainImageIds:
trainImageIds.append(string)
print("The are " + str(len(trainImageIds)) + " distinct ids." +
str(df.head()))
trainImageIds.sort()
creator = gdal_utils()
for imageId in trainImageIds:
new_img = give_image_sandwhich(imageId)
creator.create_tiff_file_from_array(
"./Data/sixteen_band/" + imageId + "_P.tif",
"./Data/image_stacks/" + imageId + ".tif", new_img)