def classify_trees(width, height, img_path):
model = get_dilated_unet(input_shape=(2048, 2048, 3),
mode='cascade',
filters=56,
n_class=3)
model.load_weights('models/trees_model_weights.hdf5')
classified_image = np.zeros(
((height // 2048 + 1) * 2048, (width // 2048 + 1) * 2048, 3))
cv2.imwrite('data/' + folder_name + '/images/trees.jpg', classified_image)
for img in os.listdir(img_path):
classified_image = cv2.imread('data/' + folder_name +
'/images/trees.jpg')
if img.endswith('.tif'):
im = Image.open(img_path + '/' + img)
x_train = np.array(im, 'f')
x_train1 = model.predict(x_train)
x = np.zeros((2048, 2048, 3))
x[np.where(x_train1[0, :, :, 0] > 0.50)] = [255, 0, 0]
x[np.where(x_train1[0, :, :, 1] > 0.50)] = [0, 255, 0]
x[np.where(x_train1[0, :, :, 2] > 0.50)] = [0, 0, 0]
stri = img[:-4].split('_')
classified_image[int(stri[2]):int(stri[2]) + 2048,
int(stri[1]):int(stri[1]) + 2048, :3] = x
cv2.imwrite('data/' + folder_name + '/images/trees.jpg',
classified_image)
del classified_image
os.remove(img_path + '/' + img)
classified_image = cv2.imread('data/' + folder_name + '/images/trees.jpg')
classified_image = classified_image[0:height, 0:width]
cv2.imwrite('data/' + folder_name + '/images/trees.jpg', classified_image)
def classify_roads(width, height, img_path, ortho):
model = get_dilated_unet(input_shape=(2048, 2048, 3),
mode='cascade',
filters=56,
n_class=3)
model.load_weights('models/roads_model_weights.hdf5')
ds = gdal.Open('data/' + folder_name + '/' + ortho)
geo_trans = ds.GetGeoTransform()
classified_image = np.zeros(
((height // 2048 + 1) * 2048, (width // 2048 + 1) * 2048, 3))
cv2.imwrite('data/' + folder_name + '/images/roads.jpg', classified_image)
for img in os.listdir(img_path):
classified_image = cv2.imread('data/' + folder_name +
'/images/roads.jpg')
if img.endswith('.tif'):
im = Image.open(img_path + '/' + img)
x_train = np.array(im, 'f')
x_train1 = model.predict(x_train)
x = np.zeros((2048, 2048, 3))
x[np.where(x_train1[0, :, :, 0] > 0.50)] = [255, 0, 0]
x[np.where(x_train1[0, :, :, 1] > 0.50)] = [0, 255, 0]
x[np.where(x_train1[0, :, :, 2] > 0.50)] = [0, 0, 0]
stri = img[:-4].split('_')
classified_image[int(stri[2]):int(stri[2]) + 2048,
int(stri[1]):int(stri[1]) + 2048, :3] = x
cv2.imwrite('data/' + folder_name + '/images/roads.jpg',
classified_image)
del classified_image
os.remove(img_path + '/' + img)
classified_image = cv2.imread('data/' + folder_name + '/images/roads.jpg')
classified_image = classified_image[0:height, 0:width]
cv2.imwrite('data/' + folder_name + '/images/roads.jpg', classified_image)
mask2poly = vector.mask.mask_to_poly_geojson(classified_image,
channel_scaling=[1, -1, -1],
bg_threshold=100,
simplify=False)
out = vector.polygon.georegister_px_df(mask2poly,
affine_obj=geo_trans,
crs='epsg:32618')
out.to_file('data/' + folder_name + '/jsons/roads.json', driver='GeoJSON')
mask2poly = vector.mask.mask_to_poly_geojson(classified_image,
channel_scaling=[-1, 1, -1],
bg_threshold=100,
simplify=False)
out = vector.polygon.georegister_px_df(mask2poly,
affine_obj=geo_trans,
crs='epsg:32618')
out.to_file('data/' + folder_name + '/jsons/footpaths.json',
driver='GeoJSON')
x_batch = np.array(x_batch, np.float32) / 255.
y_batch = np.array(y_batch, np.float32) / 255.
yield x_batch, y_batch
if __name__ == '__main__':
df_train = pd.read_csv(
'Kaggle-Carvana-Image-Masking-Challenge-master/input/train_masks.csv')
ids_train = df_train['img'].map(lambda s: s.split('.')[0])
ids_train, ids_valid = train_test_split(ids_train, test_size=0.1)
model = get_dilated_unet(input_shape=(1024, 1024, 3),
mode='cascade',
filters=32,
n_class=1)
callbacks = [
EarlyStopping(monitor='val_dice_coef',
patience=10,
verbose=1,
min_delta=1e-4,
mode='max'),
ReduceLROnPlateau(monitor='val_dice_coef',
factor=0.2,
patience=5,
verbose=1,
epsilon=1e-4,
mode='max'),
ModelCheckpoint(monitor='val_dice_coef',
x_batch = np.array(x_batch, np.float32) / 255.
y_batch = np.array(y_batch, np.float32) / 255.
yield x_batch, y_batch
if __name__ == '__main__':
df_train = pd.read_csv('input/train_masks.csv')
ids_train = df_train['img'].map(lambda s: s.split('.')[0])
ids_train, ids_valid = train_test_split(ids_train, test_size=0.1)
# lr=0.0001,
model = get_dilated_unet(input_shape=(1280, 1280, 3),
mode='cascade',
filters=32,
lr=0.0001,
n_class=1)
callbacks = [
EarlyStopping(monitor='val_dice_coef',
patience=10,
verbose=1,
min_delta=1e-4,
mode='max'),
ReduceLROnPlateau(monitor='val_dice_coef',
factor=0.2,
patience=5,
verbose=1,
epsilon=1e-4,
mode='max'),