def convert_segmentation_images(input_path, output_path, classes_array):
filenames = os.listdir(str(data_folder / input_path))
output_folder = str(data_folder / output_path)
verify_folder_exists(output_folder)
i = 0
for filename in filenames:
# In BGR format
img = cv2.imread(str(data_folder / input_path / filename))
output_img = np.zeros(img.shape)
# For each pixel:
# If pixel.color in road colors: set pixel to (0,0,0)
# If pixel.color in lane marking colors: set pixel to (1,0,0)
# Else: set pixel to (2,0,0)
for j, class_colors in enumerate(classes_array):
for color in class_colors:
mask = (img == color[::-1]).all(axis=2)
output_img[mask] = [j + 1, 0, 0]
write_path = str(Path(data_folder) / output_path / filename)
cv2.imwrite(write_path, output_img)
i += 1
if i % 100 == 0:
print("Progress: ", i, " of ", len(filenames))
def convert_carla_depth_imgs(input_path, output_path):
filenames = os.listdir(str(data_folder / input_path))
output_folder = str(data_folder / output_path)
verify_folder_exists(output_folder)
i = 0
for filename in filenames:
img = cv2.imread(str(data_folder / input_path / filename))
# Carla data is logarithmic. Reverse their algorithm: https://github.com/carla-simulator/driving-benchmarks/blob/master/version084/carla/image_converter.py
normalized_img = np.exp(
((img / 255) - np.ones(img.shape)) * 5.70378) * 255
normalized_img[:, :, 0] = 255 - normalized_img[:, :, 0]
normalized_img[:, :, 1] = 0
normalized_img[:, :, 2] = 0
write_path = str(Path(data_folder) / output_path / filename)
cv2.imwrite(write_path, normalized_img)
i += 1
if i % 100 == 0:
print("Progress: ", i, " of ", len(filenames))
def copy_cityscapes_files(input_path, output_path, only_matching=None):
verify_folder_exists(output_path)
input_path = Path(input_path)
for folder in os.listdir(input_path):
for filename in os.listdir(str(input_path / folder)):
if only_matching is not None:
if not only_matching in filename:
continue
output_name = str(output_path) + "/" + str(folder + str(filename))
copyfile(str(input_path / folder / filename), output_name)
def segment_images_psp(path_to_checkpoint, folder_name):
output_folder_name = folder_name + "_segm"
checkpoint = model_from_checkpoint_path(path_to_checkpoint)
#checkpoint = pspnet_101_cityscapes()
output_folder = data_folder / output_folder_name
verify_folder_exists(output_folder)
for filename in os.listdir(str(data_folder / folder_name)):
out_name = str(output_folder / filename).replace("jpg", "png")
in_name = str(data_folder / folder_name / filename)
checkpoint.predict_segmentation(out_fname=out_name, inp=in_name)
def visualize_segm_depth(path_to_checkpoint, folder_name):
output_segm_folder_name = folder_name + "_depth_segm"
checkpoint = unet_model_from_checkpoint_path(path_to_checkpoint)
plot_model(checkpoint, "model.png")
print(checkpoint.summary())
output_segm_folder = data_folder / output_segm_folder_name
verify_folder_exists(output_segm_folder)
for filename in os.listdir(str(data_folder / folder_name)):
out_segm_name = str(output_segm_folder / filename).replace(
"jpg", "png")
in_name = str(data_folder / folder_name / filename)
checkpoint.predict_segmentation(out_fname=out_segm_name, inp=in_name)