def preprocess_docker(algorithm_info, to_verify_list):
"""starts docker command and updates parameters appropriately"""
image_name = algorithm_info[lookup.DOCKER_IMAGE]
cmd = lookup.get_verify_cmd(algorithm_info)
volumes = {}
if lookup.VERIFY_TXT_FILE in cmd:
asset_directory = abspath(
lookup.get_algo_asset_dirs()[lookup.embeddor])
volumes[asset_directory] = {'bind': lookup.asset_dir, 'mode': 'rw'}
for to_verify in to_verify_list:
to_verify[lookup.VERIFY_TXT_FILE] = generator.generate_verify_file(
algorithm_info, to_verify)
for to_verify in to_verify_list:
assert (lookup.INPUT_IMAGE_PATH in to_verify)
original_input_path = to_verify[lookup.INPUT_IMAGE_PATH]
original_input_path = abspath(original_input_path)
local_input_dir = fs.get_directory(original_input_path)
volumes[local_input_dir] = {'bind': lookup.input_dir, 'mode': 'rw'}
input_filename = fs.get_filename(original_input_path)
new_input_path = join(lookup.input_dir, input_filename)
to_verify[lookup.INPUT_IMAGE_PATH] = new_input_path
container_id = runner.start_docker(image_name, volumes=volumes)
for to_verify in to_verify_list:
to_verify[lookup.container_id] = container_id
return [], to_verify_list
def generate_result_file(algorithm_info, to_detect, file_type, temp=False):
output_file_name = algorithm_info[lookup.uuid_descriptor]
cmd = lookup.get_cmd(algorithm_info)
if algorithm_info[
lookup.
COMMAND_TYPE] == lookup.DOCKER and lookup.PIPE_OUTPUT not in algorithm_info: #hacky fix for piping output
output_directory = lookup.result_dir
else:
output_directory = abspath(
lookup.get_algo_asset_dirs()[lookup.detector])
if lookup.INPUT_IMAGE_PATH in cmd:
output_file_name += '_' + fs.get_filename(
to_detect[lookup.INPUT_IMAGE_PATH], extension=False)
elif lookup.INPUT_IMAGE_DIRECTORY in cmd:
output_file_name += '_' + fs.get_filename(
to_detect[lookup.INPUT_IMAGE_DIRECTORY])
if temp:
output_file_name += '-temp'
output_file_name += '.' + file_type
return join(output_directory, output_file_name)
def terimination_docker(algorithm_info, verified_list):
termination_cmds = []
cmd = lookup.get_verify_cmd(algorithm_info)
docker_containers = list(
set(
list(
map(lambda verified: verified[lookup.container_id],
verified_list))))
for container_id in docker_containers:
termination_cmds.append({
lookup.COMMAND_TYPE: lookup.END_DOCKER,
lookup.COMMAND: [container_id]
})
for verified in verified_list:
asset_file_name = fs.get_filename(
generator.generate_verify_file(algorithm_info, verified))
asset_directory = lookup.get_algo_asset_dirs()[algorithm_info[
lookup.ALGORITHM_TYPE]]
old_asset_file_path = join(asset_directory, asset_file_name)
removal_cmd = ' '.join([lookup.removal_prefix, old_asset_file_path])
termination_cmds.append({
lookup.COMMAND_TYPE: lookup.NATIVE,
lookup.COMMAND: [removal_cmd]
})
return termination_cmds
def compile_txt_results(algorithm_info, source_db):
algorithm_uuid = algorithm_info[lookup.uuid_descriptor]
asset_dir = abspath(lookup.get_algo_asset_dirs()[lookup.detector])
image_files = lookup.get_image_list(source_db)
result_file_func = lambda file: join(asset_dir, algorithm_uuid + '_' + fs.get_filename(file[lookup.file_path], extension=False) + '.txt')
result_files = [{lookup.file_path: result_file_func(file), lookup.label: file[lookup.label]} for file in image_files]
results = []
for idx, result_file_info in enumerate(result_files):
file_result = fs.read_txt_file(result_file_info[lookup.file_path])
file_result = ''.join(file_result)
result = None
if algorithm_info[lookup.DETECTOR_TYPE] == lookup.binary_detector:
yes_filter = algorithm_info[lookup.regex_filter_yes]
no_filter = algorithm_info[lookup.regex_filter_no]
stego = re.search(yes_filter, file_result)
cover = re.search(no_filter, file_result)
assert (stego or cover and not (stego and cover))
if stego:
result = lookup.stego
else:
result = lookup.cover
else:
result = float(file_result)
result_file_info.update({lookup.result: result})
results.append(result_file_info)
return results
def calculate_statistics_threshold(detector_results):
"""calculates all the relevant analyzer statistics"""
metrics = collections.OrderedDict()
labels = np.array(
list(
map(lambda d: 1
if d[lookup.label] == lookup.stego else 0, detector_results)))
prediction_single_classes = np.array(
list(map(lambda d: float(d[lookup.result]), detector_results)))
prediction_both_classes = np.array(
list(
map(
lambda d:
(float(d[lookup.result]), 1.0 - float(d[lookup.result])),
detector_results)))
if len(set(labels)) == 1:
print('roc values require at least 2 labels in the test dataset')
return {lookup.result_metric: metrics}
auc_score = roc_auc_score(labels, prediction_single_classes)
ap_score = average_precision_score(labels, prediction_single_classes)
skplt.metrics.plot_roc(labels, prediction_both_classes)
detector_result_dir = lookup.get_algo_asset_dirs()[lookup.detector]
roc_curve_name = fs.get_uuid() + '-roc.png'
roc_curve_path = abspath(join(detector_result_dir, roc_curve_name))
plt.savefig(roc_curve_path, bbox_inches='tight')
metrics[lookup.roc_auc] = auc_score
metrics[lookup.ap_score] = ap_score
metrics[lookup.roc_curve] = roc_curve_path
return {lookup.result_metric: metrics}
def compile_csv_results(algorithm_info, source_db): algorithm_uuid = algorithm_info[lookup.uuid_descriptor] asset_dir = abspath(lookup.get_algo_asset_dirs()[lookup.detector]) image_files = lookup.get_image_list(source_db) image_filepath = [abspath(file[lookup.file_path]) for file in image_files] result_files = [algorithm_uuid + '_' + fs.get_filename(file[lookup.file_path], extension=False) + '.csv' for file in image_files] result_files = [join(asset_dir, result_file) for result_file in result_files] raise NotImplementedError
def generate_verify_file(algorithm_info, to_verify):
command_type = algorithm_info[lookup.COMMAND_TYPE]
if command_type == lookup.DOCKER:
file_dir = lookup.asset_dir
else:
file_dir = abspath(lookup.get_algo_asset_dirs()[lookup.embeddor])
file_name = algorithm_info[lookup.uuid_descriptor] + '_' + fs.get_filename(
to_verify[lookup.INPUT_IMAGE_PATH], extension=False) + '.txt'
file_path = join(file_dir, file_name)
return file_path
def compile_csv_directory(algorithm_info, source_db):
algorithm_uuid = algorithm_info[lookup.uuid_descriptor]
asset_dir = abspath(lookup.get_algo_asset_dirs()[lookup.detector])
image_list = lookup.get_image_list(source_db)
directory = list(set([fs.get_directory(image[lookup.file_path]) for image in image_list]))
assert(len(directory) == 1)
directory = directory[0]
result_csv_file = algorithm_uuid + '_' + fs.get_filename(directory) + '.csv'
result_csv_file = join(asset_dir, result_csv_file)
data = fs.read_csv_file(result_csv_file)
results = []
def get_image_info(file_name):
transform = lambda img: img[lookup.file_path]
if lookup.OUTPUT_FILE in algorithm_info:
if algorithm_info[lookup.OUTPUT_FILE] == lookup.INPUT_IMAGE_NAME:
transform = lambda img: fs.get_filename(img[lookup.file_path])
filtered_list = list(filter(lambda img: transform(img) == file_name, image_list))
assert(len(filtered_list) == 1)
return filtered_list[0]
for result in data:
result_info = get_image_info(result[0])
file_result = result[1]
if algorithm_info[lookup.DETECTOR_TYPE] == lookup.binary_detector:
yes_filter = algorithm_info[lookup.regex_filter_yes]
no_filter = algorithm_info[lookup.regex_filter_no]
stego = re.search(yes_filter, file_result)
cover = re.search(no_filter, file_result)
assert (stego or cover and not (stego and cover))
if stego:
result = lookup.stego
else:
result = lookup.cover
else:
result = float(file_result)
result_info.update({lookup.result: result})
results.append(result_info)
return results
def verify_embedding(verify_db, embeddors):
embeddor_results = defaultdict(list)
image_files = lookup.get_image_list(verify_db)
for image_file in image_files:
image_file[lookup.INPUT_IMAGE_PATH] = image_file[lookup.file_path]
embeddor_uuid = image_file[lookup.uuid_descriptor]
verify_txt_file = generator.generate_verify_file(embeddors[embeddor_uuid], image_file)
asset_file_name = fs.get_filename(verify_txt_file)
asset_directory = lookup.get_algo_asset_dirs()[lookup.embeddor]
verify_file_path = abspath(join(asset_directory, asset_file_name))
data = fs.read_txt_file(verify_file_path)
if (len(data[0])) == int(image_file[lookup.secret_txt_length]):
verification_result = True
else:
verification_result = False
embeddor_results[embeddor_uuid].append({lookup.INPUT_IMAGE_PATH: image_file[lookup.INPUT_IMAGE_PATH], lookup.result: verification_result})
return embeddor_results
def preprocess_docker(algorithm_info, to_detect_list):
"""starts docker command and updates parameters appropriately"""
image_name = algorithm_info[lookup.DOCKER_IMAGE]
cmd = lookup.get_cmd(algorithm_info)
volumes = {}
if lookup.INPUT_IMAGE_DIRECTORY in cmd:
updated_detect_list = generator.get_directories(to_detect_list)
for updated_detect in updated_detect_list:
docker_directory = '/' + fs.get_uuid()
volumes[updated_detect[lookup.INPUT_IMAGE_DIRECTORY]] = {
'bind': docker_directory,
'mode': 'rw'
}
updated_detect[lookup.INPUT_IMAGE_DIRECTORY] = docker_directory
elif lookup.INPUT_IMAGE_PATH in cmd:
for to_detect in to_detect_list:
original_input_path = to_detect[lookup.INPUT_IMAGE_PATH]
original_input_path = abspath(original_input_path)
local_input_dir = fs.get_directory(original_input_path)
volumes[local_input_dir] = {'bind': lookup.input_dir, 'mode': 'rw'}
input_filename = fs.get_filename(original_input_path)
new_input_path = join(lookup.input_dir, input_filename)
to_detect[lookup.INPUT_IMAGE_PATH] = new_input_path
result_directory = abspath(lookup.get_algo_asset_dirs()[lookup.detector])
assert (fs.dir_exists(result_directory))
volumes[result_directory] = {'bind': lookup.result_dir, 'mode': 'rw'}
container_id = runner.start_docker(image_name, volumes=volumes)
for to_detect in to_detect_list:
to_detect[lookup.container_id] = container_id
return [], to_detect_list