def augment(self, batch_input):
batch = []
for sample in batch_input:
aug_func = random.choice([0, 1, 1, 2, 3, 3])
if aug_func == 0:
aug_sample = sample.numpy()
elif aug_func == 1:
aug_sample = distort(sample.numpy(), 4)
elif aug_func == 2:
aug_sample = stretch(sample.numpy(), 4)
elif aug_func == 3:
aug_sample = perspective(sample.numpy())
batch.append(aug_sample)
return np.array(batch)
def augmentation(data_input, output):
"""
Run augmentor for text augmentation
"""
lines = open('{}/labels.txt'.format(data_input), 'r').readlines()
image_dirs = os.path.join(output, 'images')
imgs = os.path.join(data_input, 'images')
check_exits(image_dirs)
augment = [2, 3, 4, 5]
with open('{}/labels.txt'.format(output), 'w') as f:
for line in tqdm.tqdm(lines):
distort_aug = random.choice(augment)
stretch_aug = random.choice(augment)
try:
filename, label = line.strip().split('\t', 1)
name = os.path.splitext(filename)[0]
img_path = os.path.join(imgs, filename)
im = cv2.imread(img_path)
im = cv2.resize(im, (32 * 3, 32))
for i in range(20):
distort_img = distort(im, distort_aug)
cv2.imwrite(
'{}/distort_{}_{}.jpg'.format(image_dirs, name, i),
distort_img)
f.write('distort_{}_{}.jpg'.format(name, i))
f.write('\t')
f.write(label)
f.write('\n')
stretch_img = stretch(im, stretch_aug)
cv2.imwrite(
'{}/stretch_{}_{}.jpg'.format(image_dirs, name, i),
stretch_img)
f.write('stretch_{}_{}.jpg'.format(name, i))
f.write('\t')
f.write(label)
f.write('\n')
perspective_img = perspective(im)
cv2.imwrite(
'{}/perspec_{}_{}.jpg'.format(image_dirs, name, i),
perspective_img)
f.write('perspec_{}_{}.jpg'.format(name, i))
f.write('\t')
f.write(label)
f.write('\n')
except Exception as e:
continue
def __getitem__(self, index):
assert index <= len(self), 'index range error'
index = self.filtered_index_list[index]
with self.env.begin(write=False) as txn:
label_key = 'label-%09d'.encode() % index
label = txn.get(label_key).decode('utf-8')
img_key = 'image-%09d'.encode() % index
imgbuf = txn.get(img_key)
buf = six.BytesIO()
buf.write(imgbuf)
buf.seek(0)
try:
if self.opt.rgb:
img = Image.open(buf).convert('RGB') # for color image
else:
img = Image.open(buf).convert('L')
except IOError:
print(f'Corrupted image for {index}')
# make dummy image and dummy label for corrupted image.
if self.opt.rgb:
img = Image.new('RGB', (self.opt.imgW, self.opt.imgH))
else:
img = Image.new('L', (self.opt.imgW, self.opt.imgH))
label = '[dummy_label]'
if not self.opt.sensitive:
label = label.lower()
# We only train and evaluate on alphanumerics (or pre-defined character set in train.py)
out_of_char = f'[^{self.opt.character}]'
label = re.sub(out_of_char, '', label)
if self.opt.data_augment:
imgArray = np.asarray(img)
if torch.rand(1) < 0.3:
imgArray = augment.distort(imgArray, random.randint(3, 8))
if torch.rand(1) < 0.3:
imgArray = augment.stretch(imgArray, random.randint(3, 8))
img = Image.fromarray(np.uint8(imgArray))
return (img, label)
def generate_variations():
file_paths = glob.glob(input_images_dir +
'/*.jpg') # search for all jpg images in the folder
counter = 1
# percorre todos os arquivos de imagem da base original
for filepath in file_paths:
filename = os.path.basename(filepath)
# verifica quais fazem parte da base de treinamento
if filename in filenames_for_training:
print(counter) # apenas para controle do tempo de execucao
index_filename = filenames_for_training.index(filename)
filename_without_extension = filename.split('.')[0]
im = cv2.imread(filepath)
# para cinco níveis (2 a 11, números primos), gera novas imagens
parameters = (2, 3, 5, 7, 11)
for i in parameters:
# gera e salva imagem distorcida
distort_img = distort(im, i)
new_filename = filename_without_extension + '-distort-' + str(
counter) + '.jpg'
cv2.imwrite(images_dir + '/' + new_filename, distort_img)
adiciona_rotulo(classes_month[index_filename], new_filename)
counter = counter + 1
# gera e salva imagem esticada
stretch_img = stretch(im, i)
new_filename = filename_without_extension + '-distort-' + str(
counter) + '.jpg'
cv2.imwrite(images_dir + '/' + new_filename, stretch_img)
adiciona_rotulo(classes_month[index_filename], new_filename)
counter = counter + 1
# gera e salva imagem em perspectiva
perspective_img = perspective(im)
new_filename = filename_without_extension + '-perspective-' + str(
counter) + '.jpg'
cv2.imwrite(images_dir + '/' + new_filename, perspective_img)
adiciona_rotulo(classes_month[index_filename], new_filename)
counter = counter + 1
for image in image_list:
frames.append(image)
imageio.mimsave(gif_name, frames, 'GIF', duration=duration)
return
if __name__ == '__main__':
im = cv2.imread("imgs/demo.png")
im = cv2.resize(im, (200, 64))
cv2.imshow("im_CV", im)
distort_img_list = list()
stretch_img_list = list()
perspective_img_list = list()
for i in range(12):
distort_img = distort(im, 4)
distort_img_list.append(distort_img)
cv2.imshow("distort_img", distort_img)
stretch_img = stretch(im, 4)
cv2.imshow("stretch_img", stretch_img)
stretch_img_list.append(stretch_img)
perspective_img = perspective(im)
cv2.imshow("perspective_img", perspective_img)
perspective_img_list.append(perspective_img)
cv2.waitKey(100)
create_gif(distort_img_list, r'imgs/distort.gif')
create_gif(stretch_img_list, r'imgs/stretch.gif')
create_gif(perspective_img_list, r'imgs/perspective.gif')