'''
Functions for handling labels, filepaths and images.
'''
import os.path
import glob
from PIL import Image
from keras.utils import np_utils
import inflection
import numpy as np
import re
from config import config
class FilepathPreprocessor:
@staticmethod
def create_labels(train_path):
paths = glob.glob("{0}/*".format(train_path))
labels_file_path = config['labels_path']
with open(labels_file_path, 'w+') as labels_file:
for i, path in enumerate(paths):
for path_tuple in os.walk(path):
root = path_tuple[0]
files = path_tuple[2]
if len(files) != 0:
label = os.path.relpath(root, start=train_path)
label_number = i
image_files = []
for filename in files:
path, ext = os.path.splitext(filename)
if ext == '.png' or ext == '.jpg' or ext == '.ppm':
image_size = "_{0}".format(config['image_size'])
if image_size not in path:
image_files.append(os.path.join(label, filename))
for file in image_files:
file_line = file + " " + str(i) + "\n"
labels_file.write(file_line)
@staticmethod
def process_filepaths(paths, root_paths):
new_paths = []
for path in paths:
if path[:1] == '/':
path = path[1:]
if "_val_" in path:
full_path = os.path.join(root_paths[1], path)
else:
full_path = os.path.join(root_paths[0], path)
full_path = full_path.replace("\\", "/")
if os.path.isfile(full_path) and os.path.getsize(full_path) > 0:
new_paths.append(full_path)
# filenames for training images have a leading slash,
# which causes problems on windows with os.path.join
return new_paths
@staticmethod
def change_filepaths_after_resize(paths):
resize_paths = []
for path in paths:
image_size = "_{0}".format(config['image_size'])
if image_size not in path:
name, ext = os.path.splitext(path)
path = name + image_size + ext
resize_paths.append(path)
return resize_paths
class ImagePreprocessor:
@staticmethod
def resize_images(paths):
for path in paths:
if os.path.isfile(path) and os.path.getsize(path) > 0:
filename, ext = os.path.splitext(path)
image_size = "_{0}".format(config['image_size'])
new_filename = filename + image_size + ext
if not os.path.exists(new_filename) or os.path.getsize(new_filename) == 0:
# if the file hasn't been resized
# or the resized version is corrupt (i.e. zero size)
if image_size not in filename:
try:
image = Image.open(path)
image = image.resize(config['image_size_tuple'])
image.save(filename + image_size + ext)
except OSError:
print("OSError caused by file at {0}".format(path))
continue
# if OSError:
# cannot identify image file occurs despite above checks, skip the image
@staticmethod
def colour_images(paths):
for path in paths:
if os.path.isfile(path) and os.path.getsize(path) > 0:
filename, ext = os.path.splitext(path)
image_size = "_{0}".format(config['image_size'])
if image_size in filename:
try:
image = Image.open(path)
if image.mode != "RGB":
image = image.convert("RGB")
image.save(filename + ext)
except OSError:
print("OSError caused by file at {0}".format(path))
continue
# if OSError:
# cannot identify image file occurs despite above checks, skip the image
@staticmethod
def normalise(image_data):
image_data = image_data.astype('float')
image_data /= 255.0
image_data -= 0.5
image_data *= 2.0
return image_data
class LabelProcessor:
@staticmethod
def depluralise_string(string):
singular = inflection.singularize(string)
return singular
@staticmethod
def conflate_labels(image_label):
conflated_label_words = []
for word in image_label:
conflated_label_words.append(LabelProcessor.depluralise_string(word))
return conflated_label_words
@staticmethod
def read_categories(path):
labels_to_label_names = {}
with open(path, 'r') as categories_file:
for line in categories_file:
label_name, label = line.split(" ")
label_name = label_name.replace("_", " ")
label_name = label_name.replace("/", " ")
labels_to_label_names[int(label)] = label_name
return labels_to_label_names
@staticmethod
def parse_label_from_filename(filename):
filename_parts = re.findall(r"[a-zA-Z_]+", filename)
label_parts = []
for part in filename_parts:
filetypes = ["jpg", "png", "ppm"]
if all(file_ext not in part for file_ext in filetypes):
label_parts.append(part)
return "/".join(label_parts)
@staticmethod
def create_categories_file(labels_file_path):
filenames, labels = LabelProcessor.read_labels([labels_file_path])
category_labels = zip(filenames, labels)
seen_labels = set()
for category_label in category_labels:
filename, label = category_label
if label not in seen_labels:
filename = LabelProcessor.parse_label_from_filename(filename)
with open(config['categories_path'], "a") as categories_file:
categories_file.write(filename + " " + str(label) + "\n")
seen_labels.add(label)
@staticmethod
def read_labels(paths):
labels = []
filenames = []
for path in paths:
with open(path, 'r') as label_file:
for line in label_file:
filename, label = line.split(" ")
labels.append(int(label.rstrip('\r\n')))
filenames.append(filename)
return filenames, labels
@staticmethod
def convert_to_one_hot(labels):
return np_utils.to_categorical(labels)
@staticmethod
def convert_labels_to_label_names(labels):
chosen_label_names = []
for image_labels in labels:
labels_to_label_names = LabelProcessor.read_categories(config['categories_path'])
label_names = [labels_to_label_names[label] for label in image_labels]
if not label_names:
label_names = [""]
chosen_label_names.append(label_names)
return chosen_label_names
