def load(self, output_dir="output"):
"""Load dataset from `dataset_path` into memory and return
a tuple of (images, labels) where `image[i]` is the i'th preprocessed image
normalized into 0-1 range
and `labels[i]` is the label of i'th image.
`dataset_path` must have directories representing labels and
images inside each directory."""
cache_file = os.path.join(output_dir, self.cache_file_name)
Printer.information("Searching for cache file: " + cache_file)
if os.path.exists(cache_file):
Printer.information("Cache file found. Loading from cache.")
with open(cache_file, 'rb') as fr:
images, labels = pickle.load(fr)
else:
Printer.warning("Cache file not found")
Printer.information("Started loading dataset")
images = []
labels = []
image_paths = self._get_image_paths()
for ind, image_path in enumerate(image_paths):
image = self._process_image(image_path)
label = image_path.split(os.path.sep)[-2]
images.append(image)
labels.append(label)
Printer.processing(f"Loaded {ind}/{len(image_paths)} images.")
Printer.end_processing()
images = np.array(images, dtype='float')
labels = np.array(labels)
if not sys.getsizeof(images) > 1024 * 1024 * 1024:
with open(cache_file, 'wb') as fw:
pickle.dump((images, labels), fw)
Printer.information("Dataset loaded into memory")
self.images = images
self.labels = labels
def _get_image_paths(self):
"""Return a list of image files inside the `base_path` """
image_paths = []
for (dir_path, _, file_names) in os.walk(self.base_path):
for file_name in file_names:
if os.extsep not in file_name:
Printer.warning(
"Files without extension found: {}".format(file_name))
continue
extension = file_name.split(os.extsep)[-1]
if extension not in self.image_extensions:
Printer.warning(
"Non-image files found: {}".format(file_name))
continue
if '.ipynb' in dir_path:
Printer.warning("IPyNb caches found: {}".format(file_name))
continue
image_paths.append(os.path.join(dir_path, file_name))
random.shuffle(image_paths)
Printer.information(f"Found {len(image_paths)} images")
return image_paths
def encode_labels(self):
label_binarizer = LabelBinarizer()
self.encoded_labels = label_binarizer.fit_transform(self.labels)
self.class_names = label_binarizer.classes_
Printer.information("Labels Found: " + (", ".join(self.class_names)))
def save(self, model_output_path: str):
self.model.save(model_output_path)
Printer.information(f"Model saved as {model_output_path}")
