def __init__(self,
image_data,
data_name,
dataset_name,
img_dims):
"""
:param image_data: image data to be written to CSV file
:param data_name: unique name used to identify extracted image data
:param dataset_name: image dataset name from which to read images
:param img_dims: size for dataset images (all sizes should be the same)
"""
# Store parameters
self.image_data = image_data
self.data_name = data_name
self.dataset_name = dataset_name
self.img_dims = img_dims
self.metadata_writer = JSONWriter(self.data_name, dataset_metafile_path)
# Read image labels text file and store it
self.labels_path = base_training_data_store_path+dataset_name+'/labels.txt'
self.path_reader = TXTReader(self.labels_path)
self.path_reader.read_raw()
self.path_reader.parse()
self.labels = self.path_reader.parsed_data
# Store path where to save output image data with labels and define CSVWriter object
self.data_dir = workspace_dir+'user/training_datasets/'+self.dataset_name+'/'
self.csv_writer = CSVWriter(self.data_dir+'data.csv')
# Create storing folder for output image data
mkdir(self.data_dir)
# Store number of classes
self.n_classes = len(unique(self.labels))
def write_unclassified_dataset_paths(dataset_path, dataset_name):
dataset_dir = workspace_dir + 'user/unclassified_datasets/' + dataset_name
mkdir(dataset_dir)
images = os.listdir(dataset_path)
n_images = len(images)
writing_progress = bar.Bar("Writing image paths: ", max=n_images)
with open(dataset_dir + '/paths.txt', 'a') as paths:
for image in images:
paths.write(dataset_path + '/' + image + '\n')
writing_progress.next()
def write_training_dataset_paths(dataset_path, dataset_name):
dataset_dir = workspace_dir + 'user/training_datasets/' + dataset_name
mkdir(dataset_dir)
folders = os.listdir(dataset_path)
n_images = 0
for folder in folders:
n_images += len(os.listdir(dataset_path + '/' + folder))
with open(dataset_dir + '/paths.txt', 'a') as paths:
writing_progress = bar.Bar("Writing image paths: ", max=n_images)
for folder in folders:
images = os.listdir(dataset_path + '/' + folder)
for tp in images:
paths.write(dataset_path + '/' + folder + '/' + tp + '\n')
writing_progress.next()
def __init__(self,
image_data,
data_name,
dataset_name,
img_dims):
self.image_data = image_data
self.data_name = data_name
self.dataset_name = dataset_name
self.metadata_writer = JSONWriter(self.data_name, dataset_metafile_path)
self.img_dims = img_dims
self.data_dir = base_unclassified_data_store_path+self.dataset_name+'/'
self.csv_writer = CSVWriter(self.data_dir+'data.csv')
mkdir(self.data_dir)
self.n_classes = 0
def write_labels(dataset_path, dataset_name):
dataset_dir = workspace_dir + 'user/training_datasets/' + dataset_name
mkdir(dataset_dir)
folders = os.listdir(dataset_path)
n_images = 0
for folder in folders:
n_images += len(os.listdir(dataset_path + '/' + folder))
class_id = {}
with open(dataset_dir + '/labels.txt', 'a') as labels:
writing_progress = bar.Bar("Writing image labels: ", max=n_images)
for i, folder_name in enumerate(folders):
for image in os.listdir(dataset_path + '/' + folder_name):
labels.write(folder_name + '\n')
writing_progress.next()
class_id[i] = folder_name
with open(dataset_dir + '/class_id.json', 'w') as txt:
json.dump(class_id, txt, indent=3, sort_keys=True)
def write_predictions(self):
mkdir(base_predictions_store_path)
mkdir(base_predictions_store_path + '/' + self.model_name)
mkdir(base_predictions_store_path + '/' + self.model_name + '/' +
self.classification_name)
for image_n in range(self.n_images):
self.X_[image_n] = np.append(
self.X_[image_n], self.predictions[self.image_paths[image_n]])
with open(self.raw_predictions_path, 'w') as csvfile:
writer = csv.writer(csvfile, delimiter=',')
for image_prediction in self.X_:
writer.writerow(image_prediction)
with open(self.predictions_paths_path, 'w') as pathfile:
writer = csv.writer(pathfile, delimiter=',')
for n in range(self.n_images):
writer.writerow([
self.image_paths[n], self.predictions[self.image_paths[n]]
])
self.image_paths = np.asarray(self.image_paths)
np.random.shuffle(self.image_paths)
if self.show_images:
for n in range(self.max_images):
display_image(self.predictions[self.image_paths[n]],
self.image_paths[n])