def __init__(self, cfg, unique_id, apply_threshold, is_bert):
self.dataset_cfg = get_dataset_metadata_cfg()
self.workspace_dir = os.path.join(cfg["workspace"]["directory"],
cfg["dataset"]["name"])
self.tokenized_descriptions_file_path = os.path.join(
self.workspace_dir, tokenized_descriptions)
self.word_dictionary_file_path = os.path.join(self.workspace_dir,
word_dictionary)
self.vector_encoding_file_path = os.path.join(self.workspace_dir,
vector_encoding)
self.processed_descriptions_file_path = os.path.join(
self.workspace_dir, index_replaced_file)
self.index_file = self.dataset_cfg["data"]["googlecc"]["index_file"]
self.dataset_path = self.dataset_cfg["data"]["googlecc"][
"dataset_path"]
self.cfg = cfg
self.unique_id = unique_id
self.is_bert = is_bert
# Prepare the dataset processing original descriptions to new descriptions
self.prepare_dataset()
self.train_test_split()
if is_bert:
BERT_MODEL_HUB = "https://tfhub.dev/google/bert_uncased_L-12_H-768_A-12/1"
with tf.Graph().as_default():
bert_module = hub.Module(BERT_MODEL_HUB)
tokenization_info = bert_module(signature="tokenization_info",
as_dict=True)
with tf.compat.v1.Session() as sess:
vocab_file, do_lower_case = tokenization_info[
"vocab_file"], tokenization_info["do_lower_case"]
print("vocab", vocab_file)
tokenizer = tokenization.FullTokenizer(vocab_file, do_lower_case)
tokenize_descriptions_bert(self.processed_descriptions_file_path,
self.tokenized_descriptions_file_path,
tokenizer)
vector_encode_descriptions_bert(
self.tokenized_descriptions_file_path,
self.vector_encoding_file_path, tokenizer)
else:
if apply_threshold:
tokenize_descriptions_with_threshold(
self.processed_descriptions_file_path,
self.tokenized_descriptions_file_path)
else:
tokenize_descriptions(self.processed_descriptions_file_path,
self.tokenized_descriptions_file_path)
create_word_map(self.tokenized_descriptions_file_path,
self.word_dictionary_file_path)
word_map = read_word_dictionary(self.word_dictionary_file_path)
vector_encode_descriptions(self.tokenized_descriptions_file_path,
self.vector_encoding_file_path,
word_map)
def retrieve_data_list_file(cfg, run_type):
dataset_cfg = get_dataset_metadata_cfg()
if run_type == "train":
data_list = dataset_cfg["data"]["flickr"]["train_images_file"]
elif run_type is "validation":
data_list = dataset_cfg["data"]["flickr"]["validation_images_file"]
else:
data_list = dataset_cfg["data"]["flickr"]["test_images_file"]
return data_list
def image_generator(cfg, data_list):
dataset_metadata = get_dataset_metadata_cfg()
with open(data_list, 'r') as file:
for line in file:
image = cv2.imread(
os.path.join(
dataset_metadata["data"]["flickr"]["dataset_path"],
line.strip()))
image = np.expand_dims(np.asarray(cv2.resize(image,
(299, 299))) / 255.0,
axis=0)
yield image
def __init__(self, cfg):
dataset_cfg = get_dataset_metadata_cfg()
self.workspace_dir = cfg["workspace"]["directory"]
self.tokenized_descriptions_file_path = os.path.join(
self.workspace_dir, tokenized_descriptions)
self.word_dictionary_file_path = os.path.join(self.workspace_dir,
word_dictionary)
self.vector_encoding_file_path = os.path.join(self.workspace_dir,
vector_encoding)
self.cfg = cfg
tokenize_descriptions(dataset_cfg["data"]["flickr"]["descriptions"],
self.tokenized_descriptions_file_path)
create_word_map(self.tokenized_descriptions_file_path,
self.word_dictionary_file_path)
word_map = read_word_dictionary(self.word_dictionary_file_path)
vector_encode_descriptions(self.tokenized_descriptions_file_path,
self.vector_encoding_file_path, word_map)
def __init__(self, cfg, unique_id, shape):
self.dataset_cfg = get_dataset_metadata_cfg()
self.workspace_dir = os.path.join(cfg["workspace"]["directory"],
cfg["dataset"]["name"])
self.tokenized_descriptions_file_path = os.path.join(
self.workspace_dir, tokenized_descriptions)
self.word_dictionary_file_path = os.path.join(self.workspace_dir,
word_dictionary)
self.vector_encoding_file_path = os.path.join(self.workspace_dir,
vector_encoding)
self.processed_descriptions_file_path = os.path.join(
self.workspace_dir, index_replaced_file)
self.index_file = self.dataset_cfg["data"]["googlecc"]["index_file"]
self.dataset_path = self.dataset_cfg["data"]["googlecc"][
"dataset_path"]
self.apply_dataset_threshold = cfg["dataset"]["is_threshold"]
self.cfg = cfg
self.shape = shape
self.unique_id = unique_id
# Prepare the dataset processing original descriptions to new descriptions
self.prepare_dataset()
self.train_test_split()
if self.apply_dataset_threshold:
tokenize_descriptions_with_threshold(
self.processed_descriptions_file_path,
self.tokenized_descriptions_file_path)
else:
tokenize_descriptions(self.processed_descriptions_file_path,
self.tokenized_descriptions_file_path)
create_word_map(self.tokenized_descriptions_file_path,
self.word_dictionary_file_path)
word_map = read_word_dictionary(self.word_dictionary_file_path)
vector_encode_descriptions(self.tokenized_descriptions_file_path,
self.vector_encoding_file_path, word_map)
clear_session()
parser = argparse.ArgumentParser(description="config")
parser.add_argument(
"--config",
nargs="?",
type=str,
default="../configs/resnet50_lstm_with_threshold.yaml",
help="Configuration file to use",
)
args = parser.parse_args()
with open(args.config) as fp:
cfg = yaml.load(fp)
dataset_cfg = get_dataset_metadata_cfg()
model_workspace_dir = os.path.join(cfg["workspace"]["directory"],
cfg["dataset"]["name"],
cfg["model"]["arch"])
utils.make_directories(model_workspace_dir)
img_model = ResNet50(weights='imagenet')
dataset_preprocessor = PreProcessing(cfg, "resnet50", False, False)
dataset_preprocessor.run_one_time_encoding(img_model)
# Load train, validation sets from the pre-processor
training_generator, validation_generator, test_generator = dataset_preprocessor.get_keras_generators(
"resnet50")
MAX_LEN = 40