def gtiff_dsm_read_then_create_from_numpy_array():
"""
This demonstrates how to read a geotiff file in, add a bias term, update the nodata values, and write out to disk
"""
print("")
print("CREATE GEOTIFF FROM NUMPY ARRAY DEMO")
file_utils.make_dir_if_not_exists(save_dir)
gtiff_image_object = ImageFactory.geotiff.from_file(gtiff_full_path)
save_fname = "gtiff_dsm_ortho_from_numpy.tif"
numpy_save_fullpath = os.path.join(save_dir, save_fname)
gtiff_from_numpy = ImageFactory.geotiff.from_numpy_array(gtiff_image_object.read_all_image_data_from_disk(),
gtiff_image_object.get_point_calculator().get_geot(),
gtiff_image_object.get_point_calculator().get_projection(),
gtiff_image_object.get_metadata().get_nodata_val())
gtiff_from_numpy.write_to_disk(numpy_save_fullpath)
numpy_saved_image = ImageFactory.geotiff.from_file(numpy_save_fullpath)
data = numpy_saved_image.read_all_image_data_from_disk()
val_to_add = 10
data = data + val_to_add
save_fname = "gtiff_dsm_ortho_from_numpy_plus_ten.tif"
numpy_save_fullpath = os.path.join(save_dir, save_fname)
gtiff_from_numpy.set_image_data(data)
gtiff_from_numpy.get_metadata().set_nodata_val(gtiff_image_object.get_metadata().get_nodata_val() + val_to_add)
gtiff_from_numpy.write_to_disk(numpy_save_fullpath)
print("geotiff written to: " + numpy_save_fullpath)
def gtiff_dsm_image_object_save():
print("")
print("GTIFF DSM IMAGE OBJECT SAVE DEMO")
file_utils.make_dir_if_not_exists(save_dir)
gtiff_image_object = ImageFactory.geotiff.from_file(gtiff_full_path)
save_fname = "gtiff_dsm_write_to_disk.tif"
save_fullpath = os.path.join(save_dir, save_fname)
gtiff_image_object.write_to_disk(save_fullpath)
print("geotiff written to: " + save_fullpath)
def train_model(
base_model, # type: Model
training_generator, # type: training_generator
validation_generator, # type: training_generator
output_dir, # type: str
final_model_layers=None, # type: Tensor
model_weights_file=None, # type: str
model_output_fname="trained_model.h5", # type: str
n_epochs=20, # type: int
loss="categorical_crossentropy", # type: str
optimizer=optimizers.Adam(), # type: Optimizer
metrics=["accuracy"] # type: list
): # type: (...) -> None
tensorboard_log_dir = os.path.join(output_dir, "logs")
tensorboard_log_dir = os.path.join(tensorboard_log_dir,
"{}".format(time()))
file_utils.make_dir_if_not_exists(tensorboard_log_dir)
tensorboard = TensorBoard(log_dir=tensorboard_log_dir)
filepath = os.path.join(output_dir, model_output_fname)
labels_fname = os.path.join(output_dir, "labels.txt")
sorted_class_labels = sorted(list(training_generator.class_indices.keys()))
class_labels_fname = os.path.join(output_dir, "labels.txt")
file_utils.write_text_list_to_file(sorted_class_labels, class_labels_fname)
checkpoint = ModelCheckpoint(filepath,
monitor='val_loss',
verbose=1,
save_best_only=True,
save_weights_only=False,
mode='min',
period=1)
callbacks_list = [checkpoint, tensorboard]
if final_model_layers is None:
model = base_model
else:
model = Model(inputs=base_model.input, outputs=final_model_layers)
if model_weights_file is not None:
model.load_weights(model_weights_file)
model.compile(loss=loss, optimizer=optimizer, metrics=metrics)
num_training_img = len(training_generator.filenames)
num_validation_img = len(validation_generator.filenames)
steps_per_epoch = num_training_img / training_generator.batch_size
validation_steps = num_validation_img / validation_generator.batch_size
model.fit_generator(training_generator,
steps_per_epoch=steps_per_epoch,
epochs=n_epochs,
callbacks=callbacks_list,
validation_data=validation_generator,
validation_steps=validation_steps)
reflectance_image_data = reflectance_envi_image.read_all_image_data_from_disk()
reflectance_image_data = reflectance_image_data.astype(float) * 0.01
ny, nx, nbands = reflectance_image_data.shape
seaborn_color_palette = seaborn.color_palette("RdBu_r")
test_spectrum_fname = file_utils.get_path_from_subdirs(demo_data_base_dir, [
"image_data", "hyperspectral", "cooke_city", "self_test", "SPL", "F1",
"F1_l.txt"
])
spectrum = SpectrumFactory.envi.ascii.EnviAsciiPlotFileFactory.from_ascii_file(
test_spectrum_fname)
save_dir = os.path.join(demo_data_save_dir, "spectral_target_detection_demos")
file_utils.make_dir_if_not_exists(save_dir)
def save_rgb():
print("")
print("SAVE RGB DEMO")
red_band = 16
green_band = 6
blue_band = 1
save_image_fname = os.path.join(save_dir, "cooke_city_rgb.png")
rgb_image = np.zeros((ny, nx, 3))
rgb_image[:, :, 0] = reflectance_image_data[:, :, red_band]
rgb_image[:, :, 1] = reflectance_image_data[:, :, green_band]
rgb_image[:, :, 2] = reflectance_image_data[:, :, blue_band]
imageio.imwrite(save_image_fname, rgb_image)
def create_maptiles():
print("")
print("CREATE MAPTILES DEMO")
file_utils.make_dir_if_not_exists(save_dir)
gtiff_image_object = ImageFactory.geotiff.from_file(gtiff_full_path)
gtiff_image_object.create_map_tiles("/tmp/")
def create_vehicle_training_chips():
print("CHIPPING OUT VEHICLES FROM TRAINING IMAGE, THIS MIGHT TAKE A WHILE")
print("save dir = " + save_dir_base)
chipper_ny = 50
chipper_nx = 50
output_chip_ny = 244
output_chip_nx = 244
vehicle_mask = imread(png_positive_mask_fullpath)
not_vehicle_mask = imread(png_negative_mask_fullpath)
image_data = imread(png_fullpath)
vehicle_locs_y, vehicle_locs_x = image_mask_utils.rgb_image_point_mask_to_pixel_locations(
vehicle_mask)
# remove chips that aren't very vehicle like, or have vehicles that are badly obscured.
vehicle_yx_to_remove = [(1547, 524), (1813, 725), (1841, 708), (2135, 674)]
vehicle_loc_tuples = numpy_and_array_utils.separate_lists_to_list_of_tuples(
[vehicle_locs_y, vehicle_locs_x])
for yx_to_remove in vehicle_yx_to_remove:
vehicle_loc_tuples.remove(yx_to_remove)
vehicle_locs_y, vehicle_locs_x = numpy_and_array_utils.list_of_tuples_to_separate_lists(
vehicle_loc_tuples)
# chip out the vehicles
vehicle_chips, vehicle_upper_lefts = image_chipper.\
chip_images_by_pixel_centers(image_data,
vehicle_locs_y,
vehicle_locs_x,
chip_ny_pixels=chipper_ny,
chip_nx_pixels=chipper_nx)
not_vehicle_locs_y, not_vehicle_locs_x = image_mask_utils.rgb_image_point_mask_to_pixel_locations(
not_vehicle_mask)
# remove chips that do actually contain vehicles.
not_vehicle_yx_to_remove = [(431, 1811), (1039, 377), (1143, 415),
(1186, 192), (1186, 193), (1190, 150)]
not_vehicle_loc_tuples = numpy_and_array_utils.separate_lists_to_list_of_tuples(
[not_vehicle_locs_y, not_vehicle_locs_x])
for yx_to_remove in not_vehicle_yx_to_remove:
not_vehicle_loc_tuples.remove(yx_to_remove)
not_vehicle_locs_y, not_vehicle_locs_x = numpy_and_array_utils.list_of_tuples_to_separate_lists(
not_vehicle_loc_tuples)
not_vehicle_chips, not_vehicle_upper_lefts = image_chipper.\
chip_images_by_pixel_centers(image_data,
not_vehicle_locs_y,
not_vehicle_locs_x,
chip_ny_pixels=chipper_ny,
chip_nx_pixels=chipper_nx)
vehicles_save_dir = os.path.join(save_dir_base, "vehicle")
not_vehicles_save_dir = os.path.join(save_dir_base, "not_vehicle")
file_utils.make_dir_if_not_exists(vehicles_save_dir)
file_utils.make_dir_if_not_exists(not_vehicles_save_dir)
# create filenames for vehicles and non-vehicles based on original center locations. This will let us
# look through our chips for false positives and remove them from the masks
vehicle_chips_fnames_list = [
"vehicle_center_y_" + str(vehicle_locs_y[i]) + "_center_x_" +
str(vehicle_locs_x[i]) for i in range(len(vehicle_locs_x))
]
not_vehicle_chips_fnames_list = [
"not_vehicle_center_y_" + str(not_vehicle_locs_y[i]) + "_center_x_" +
str(not_vehicle_locs_x[i]) for i in range(len(not_vehicle_locs_x))
]
image_chipper.write_chips_to_disk(vehicle_chips,
vehicles_save_dir,
fnames_list=vehicle_chips_fnames_list,
output_chip_ny=output_chip_ny,
output_chip_nx=output_chip_nx)
image_chipper.write_chips_to_disk(
not_vehicle_chips,
not_vehicles_save_dir,
fnames_list=not_vehicle_chips_fnames_list,
output_chip_ny=output_chip_ny,
output_chip_nx=output_chip_nx)
print("DONE CHIPPING OUT IMAGES")
def create_vehicle_training_chips():
print("CHIPPING OUT VEHICLES FROM EVALUATION IMAGE, THIS MIGHT TAKE A WHILE")
print("save dir = " + save_dir_base)
chipper_ny = 50
chipper_nx = 50
output_chip_ny = 244
output_chip_nx = 244
vehicle_mask = imread(png_positive_mask_fullpath)
not_vehicle_mask = imread(png_negative_mask_fullpath)
image_data = imread(png_fullpath)
vehicle_locs_y, vehicle_locs_x = image_mask_utils.rgb_image_point_mask_to_pixel_locations(vehicle_mask)
# remove chips that aren't very vehicle like, or have vehicles that are badly obscured.
vehicle_yx_to_remove = [(1583, 1650),
(1599, 1217),
(2094, 122),
(2106, 160)]
vehicle_loc_tuples = numpy_and_array_utils.separate_lists_to_list_of_tuples([vehicle_locs_y, vehicle_locs_x])
for yx_to_remove in vehicle_yx_to_remove:
vehicle_loc_tuples.remove(yx_to_remove)
vehicle_locs_y, vehicle_locs_x = numpy_and_array_utils.list_of_tuples_to_separate_lists(vehicle_loc_tuples)
# chip out the vehicles
vehicle_chips, vehicle_upper_lefts = image_chipper.\
chip_images_by_pixel_centers(image_data,
vehicle_locs_y,
vehicle_locs_x,
chip_ny_pixels=chipper_ny,
chip_nx_pixels=chipper_nx)
not_vehicle_locs_y, not_vehicle_locs_x = image_mask_utils.rgb_image_point_mask_to_pixel_locations(not_vehicle_mask)
# remove chips that do actually contain vehicles.
not_vehicle_yx_to_remove = [(489, 1124),
(584, 1186),
(1222, 84),
(1271, 1409),
(1278, 1421),
(1487, 1497),
(1499, 1883),
(1531, 1444),
(1597, 1820),
(1601, 1766),
(1659, 922),
(1677, 811),
(1694, 780)]
not_vehicle_loc_tuples = numpy_and_array_utils.separate_lists_to_list_of_tuples([not_vehicle_locs_y, not_vehicle_locs_x])
for yx_to_remove in not_vehicle_yx_to_remove:
not_vehicle_loc_tuples.remove(yx_to_remove)
not_vehicle_locs_y, not_vehicle_locs_x = numpy_and_array_utils.list_of_tuples_to_separate_lists(not_vehicle_loc_tuples)
not_vehicle_chips, not_vehicle_upper_lefts = image_chipper.\
chip_images_by_pixel_centers(image_data,
not_vehicle_locs_y,
not_vehicle_locs_x,
chip_ny_pixels=chipper_ny,
chip_nx_pixels=chipper_nx)
vehicles_save_dir = os.path.join(save_dir_base, "vehicle")
not_vehicles_save_dir = os.path.join(save_dir_base, "not_vehicle")
file_utils.make_dir_if_not_exists(vehicles_save_dir)
file_utils.make_dir_if_not_exists(not_vehicles_save_dir)
# create filenames for vehicles and non-vehicles based on original center locations. This will let us
# look through our chips for false positives and remove them from the masks
vehicle_chips_fnames_list = ["vehicle_center_y_" + str(vehicle_locs_y[i]) + "_center_x_" + str(vehicle_locs_x[i]) for i in range(len(vehicle_locs_x))]
not_vehicle_chips_fnames_list = ["not_vehicle_center_y_" + str(not_vehicle_locs_y[i]) + "_center_x_" + str(not_vehicle_locs_x[i]) for i in range(len(not_vehicle_locs_x))]
image_chipper.write_chips_to_disk(vehicle_chips, vehicles_save_dir,
fnames_list=vehicle_chips_fnames_list,
output_chip_ny=output_chip_ny, output_chip_nx=output_chip_nx)
image_chipper.write_chips_to_disk(not_vehicle_chips, not_vehicles_save_dir,
fnames_list=not_vehicle_chips_fnames_list,
output_chip_ny=output_chip_ny, output_chip_nx=output_chip_nx)
print("DONE CHIPPING OUT IMAGES")
