def augment_image(image_num_pair, args):
filepath, n = image_num_pair
img = cv2.imread(filepath)
filename = filepath.split(os.sep)[-1]
dot_pos = filename.rfind('.')
imgname = filename[:dot_pos]
ext = filename[dot_pos:]
print('Augmenting {} ...'.format(filename))
for i in range(n):
img_varied = img.copy()
varied_imgname = '{}_{:0>3d}_'.format(imgname, i)
if random.random() < args.p_mirror:
img_varied = cv2.flip(img_varied, 1)
varied_imgname += 'm'
if random.random() < args.p_crop:
img_varied = ia.random_crop(img_varied, args.crop_size,
args.crop_hw_vari)
varied_imgname += 'c'
if random.random() < args.p_rotate:
img_varied = ia.random_rotate(img_varied, args.rotate_angle_vari,
args.p_rotate_crop)
varied_imgname += 'r'
if random.random() < args.p_hsv:
img_varied = ia.random_hsv_transform(img_varied, args.hue_vari,
args.sat_vari, args.val_vari)
varied_imgname += 'h'
if random.random() < args.p_gamma:
img_varied = ia.random_gamma_transform(img_varied, args.gamma_vari)
varied_imgname += 'g'
output_filepath = os.sep.join(
[args.output_dir, '{}{}'.format(varied_imgname, ext)])
cv2.imwrite(output_filepath, img_varied)
def augment_images(filelist, args):
# 遍历所有列表内的文件
for filepath, n in filelist:
img = cv2.imread(filepath)
filename = filepath.split(os.sep)[-1]
dot_pos = filename.rfind('.')
# 获取文件名和后缀名
imgname = filename[:dot_pos]
ext = filename[dot_pos:]
print('Augmenting {} ...'.format(filename))
for i in range(n):
img_varied = img.copy()
# 扰动后文件名的前缀
varied_imgname = '{}_{:0>3d}_'.format(imgname, i)
# 按照比例随机对图像进行镜像
if random.random() < args.p_mirror:
# 利用numpy.fliplr(img_varied)也能实现
img_varied = cv2.flip(img_varied, 1)
varied_imgname += 'm'
# 按照比例随机对图像进行裁剪
if random.random() < args.p_crop:
img_varied = ia.random_crop(img_varied, args.crop_size,
args.crop_hw_vari)
varied_imgname += 'c'
# 按照比例随机对图像进行旋转
if random.random() < args.p_rotate:
img_varied = ia.random_rotate(img_varied,
args.rotate_angle_vari,
args.p_rotate_crop)
varied_imgname += 'r'
# 按照比例随机对图像进行HSV扰动
if random.random() < args.p_hsv:
img_varied = ia.random_hsv_transform(img_varied, args.hue_vari,
args.sat_vari,
args.val_vari)
varied_imgname += 'h'
# 按照比例随机对图像进行Gamma扰动
if random.random() < args.p_gamma:
img_varied = ia.random_gamma_transform(img_varied,
args.gamma_vari)
varied_imgname += 'g'
# 生成扰动后的文件名并保存在指定的路径
output_filepath = os.sep.join(
[args.output_dir, '{}{}'.format(varied_imgname, ext)])
cv2.imwrite(output_filepath, img_varied)
def random_augmentation(image):
if random.random() < 0.0:
return image
# 按照比例随机对图像进行镜像
img_varied = random_flip(image)
# 随机亮度
if random.random() < 0.5:
t = random.randint(-100, 100)
blank = np.zeros(image.shape, image.dtype)
img_varied = cv2.addWeighted(img_varied, 1, blank, 1, t)
# 随机对比度
if random.random() < 0.5:
t = random.randint(-100, 100)
blank = np.zeros(image.shape, image.dtype)
img_varied = cv2.addWeighted(img_varied, 1 + t / 100.0, blank,
1 - t / 100.0, 0)
# 按照比例随机对图像进行HSV扰动
if random.random() < 1:
img_varied = ia.random_hsv_transform(img_varied, 10, 0.1, 0.1)
# 按照比例随机对图像进行Gamma扰动
if random.random() < 0.5:
img_varied = ia.random_gamma_transform(img_varied, 2.0)
return img_varied
# files = glob.glob("./data/image/mouth/pos/*.png")
# random.shuffle(files)
# c_image = cv2.imread(files[0])
# cv2.imshow("o",c_image)
# imgae = random_augmentation(c_image)
# cv2.imshow("a",imgae)
# cv2.waitKey(0)
# # 按照比例随机对图像进行裁剪
# if random.random() < 1:
# img_varied = ia.random_crop(
# img_varied,
# 0.1,
# 0.8)
# # 按照比例随机对图像进行旋转
# if random.random() < 1.0:
# img_varied = ia.random_rotate(
# img_varied,
# 1.0,
# 10.0)
def augment_image(image_num_pair, args):
filepath, n = image_num_pair
img = cv2.imread(filepath)
filename = filepath.split(os.sep)[-1]
dot_pos = filename.rfind('.')
imgname = filename[:dot_pos]
ext = filename[dot_pos:]
print('Augmenting {} ...'.format(filename))
for i in range(n):
img_varied = img.copy()
varied_imgname = '{}_{:0>3d}_'.format(imgname, i)
if random.random() < args.p_mirror:
img_varied = cv2.flip(img_varied, 1)
varied_imgname += 'm'
if random.random() < args.p_crop:
img_varied = ia.random_crop(
img_varied,
args.crop_size,
args.crop_hw_vari)
varied_imgname += 'c'
if random.random() < args.p_rotate:
img_varied = ia.random_rotate(
img_varied,
args.rotate_angle_vari,
args.p_rotate_crop)
varied_imgname += 'r'
if random.random() < args.p_hsv:
img_varied = ia.random_hsv_transform(
img_varied,
args.hue_vari,
args.sat_vari,
args.val_vari)
varied_imgname += 'h'
if random.random() < args.p_gamma:
img_varied = ia.random_gamma_transform(
img_varied,
args.gamma_vari)
varied_imgname += 'g'
output_filepath = os.sep.join([
args.output_dir,
'{}{}'.format(varied_imgname, ext)])
cv2.imwrite(output_filepath, img_varied)
def augment_images(filelist, args):
for filepath, n in filelist:
filepath = filepath.replace('\\', '/')
img = cv_imread(filepath)
#print(filepath)
#print(img)
filename = filepath.split('/')[-1]
dot_pos = filename.rfind('.')
imgname = filename[:dot_pos]
ext = filename[dot_pos:]
#print(ext, imgname,filename)
print('Augmenting {} ...'.format(filename))
for i in range(n):
img_varied = img.copy()
varied_imgname = '{}_{:0>3d}_'.format(imgname, i)
if random.random() < args.p_mirror:
img_varied = cv.flip(img_varied, 1)
varied_imgname += 'm'
if random.random() < args.p_crop:
img_varied = ia.random_crop(img_varied, args.crop_size,
args.crop_hw_vari)
varied_imgname += 'c'
if random.random() < args.p_rotate:
img_varied = ia.random_rotate(img_varied,
args.rotate_angle_vari,
args.p_rotate_crop)
varied_imgname += 'r'
if random.random() < args.p_hsv:
img_varied = ia.random_hsv_transform(img_varied, args.hue_vari,
args.sat_vari,
args.val_vari)
varied_imgname += 'h'
if random.random() < args.p_gamma:
img_varied = ia.random_gamma_transform(img_varied,
args.gamma_vari)
varied_imgname += 'g'
output_filepath = os.sep.join(
[args.output_dir, '{}{}'.format(varied_imgname,
ext)]).replace('\\', '/')
#cv.imwrite(output_filepath, img_varied)
cv.imencode('.png', img_varied)[1].tofile(output_filepath)
def generator(input_size=512,
batch_size=32,
background_ratio=3. / 8,
random_scale=np.array([0.5, 1, 2.0, 3.0]),
vis=False):
image_list = np.array(get_images())
print('{} training images in {}'.format(image_list.shape[0],
FLAGS.training_data_path))
index = np.arange(0, image_list.shape[0])
while True:
np.random.shuffle(index)
images = []
image_fns = []
score_maps = []
geo_maps = []
training_masks = []
for i in index:
try:
im_fn = image_list[i]
im = cv2.imread(im_fn)
#augmentation 1
im = random_hsv_transform(im, 10, 0.5, 0.5)
# print im_fn
h, w, _ = im.shape
txt_fn = im_fn.replace(
os.path.basename(im_fn).split('.')[1], 'txt')
if not os.path.exists(txt_fn):
print('text file {} does not exists'.format(txt_fn))
continue
text_polys, text_tags = load_annoataion(txt_fn)
#augmentation 2
im, text_polys = random_flip_left_right(im, text_polys)
text_polys, text_tags = check_and_validate_polys(
text_polys, text_tags, (h, w))
# if text_polys.shape[0] == 0:
# continue
# random scale this image
rd_scale = np.random.choice(random_scale)
im = cv2.resize(im, dsize=None, fx=rd_scale, fy=rd_scale)
text_polys *= rd_scale
# print rd_scale
# random crop a area from image
if np.random.rand() < background_ratio:
# crop background
im, text_polys, text_tags = crop_area(im,
text_polys,
text_tags,
crop_background=True)
if text_polys.shape[0] > 0:
# cannot find background
continue
# pad and resize image
new_h, new_w, _ = im.shape
max_h_w_i = np.max([new_h, new_w, input_size])
im_padded = np.zeros((max_h_w_i, max_h_w_i, 3),
dtype=np.uint8)
im_padded[:new_h, :new_w, :] = im.copy()
im = cv2.resize(im_padded, dsize=(input_size, input_size))
score_map = np.zeros((input_size, input_size),
dtype=np.uint8)
geo_map_channels = 5 if FLAGS.geometry == 'RBOX' else 8
geo_map = np.zeros(
(input_size, input_size, geo_map_channels),
dtype=np.float32)
training_mask = np.ones((input_size, input_size),
dtype=np.uint8)
else:
im, text_polys, text_tags = crop_area(
im, text_polys, text_tags, crop_background=False)
if text_polys.shape[0] == 0:
continue
h, w, _ = im.shape
# pad the image to the training input size or the longer side of image
new_h, new_w, _ = im.shape
max_h_w_i = np.max([new_h, new_w, input_size])
im_padded = np.zeros((max_h_w_i, max_h_w_i, 3),
dtype=np.uint8)
im_padded[:new_h, :new_w, :] = im.copy()
im = im_padded
# resize the image to input size
new_h, new_w, _ = im.shape
resize_h = input_size
resize_w = input_size
im = cv2.resize(im, dsize=(resize_w, resize_h))
resize_ratio_3_x = resize_w / float(new_w)
resize_ratio_3_y = resize_h / float(new_h)
text_polys[:, :, 0] *= resize_ratio_3_x
text_polys[:, :, 1] *= resize_ratio_3_y
new_h, new_w, _ = im.shape
score_map, geo_map, training_mask = generate_rbox(
(new_h, new_w), text_polys, text_tags)
if vis:
fig, axs = plt.subplots(3, 2, figsize=(20, 30))
# axs[0].imshow(im[:, :, ::-1])
# axs[0].set_xticks([])
# axs[0].set_yticks([])
# for poly in text_polys:
# poly_h = min(abs(poly[3, 1] - poly[0, 1]), abs(poly[2, 1] - poly[1, 1]))
# poly_w = min(abs(poly[1, 0] - poly[0, 0]), abs(poly[2, 0] - poly[3, 0]))
# axs[0].add_artist(Patches.Polygon(
# poly * 4, facecolor='none', edgecolor='green', linewidth=2, linestyle='-', fill=True))
# axs[0].text(poly[0, 0] * 4, poly[0, 1] * 4, '{:.0f}-{:.0f}'.format(poly_h * 4, poly_w * 4),
# color='purple')
# axs[1].imshow(score_map)
# axs[1].set_xticks([])
# axs[1].set_yticks([])
axs[0, 0].imshow(im[:, :, ::-1])
axs[0, 0].set_xticks([])
axs[0, 0].set_yticks([])
for poly in text_polys:
poly_h = min(abs(poly[3, 1] - poly[0, 1]),
abs(poly[2, 1] - poly[1, 1]))
poly_w = min(abs(poly[1, 0] - poly[0, 0]),
abs(poly[2, 0] - poly[3, 0]))
axs[0, 0].add_artist(
Patches.Polygon(poly,
facecolor='none',
edgecolor='green',
linewidth=2,
linestyle='-',
fill=True))
axs[0, 0].text(poly[0, 0],
poly[0, 1],
'{:.0f}-{:.0f}'.format(poly_h, poly_w),
color='purple')
axs[0, 1].imshow(score_map[::, ::])
axs[0, 1].set_xticks([])
axs[0, 1].set_yticks([])
axs[1, 0].imshow(geo_map[::, ::, 0])
axs[1, 0].set_xticks([])
axs[1, 0].set_yticks([])
axs[1, 1].imshow(geo_map[::, ::, 1])
axs[1, 1].set_xticks([])
axs[1, 1].set_yticks([])
axs[2, 0].imshow(geo_map[::, ::, 2])
axs[2, 0].set_xticks([])
axs[2, 0].set_yticks([])
axs[2, 1].imshow(training_mask[::, ::])
axs[2, 1].set_xticks([])
axs[2, 1].set_yticks([])
plt.tight_layout()
plt.show()
plt.close()
images.append(im[:, :, ::-1].astype(np.float32))
image_fns.append(im_fn)
score_maps.append(score_map[::4, ::4,
np.newaxis].astype(np.float32))
geo_maps.append(geo_map[::4, ::4, :].astype(np.float32))
training_masks.append(
training_mask[::4, ::4, np.newaxis].astype(np.float32))
if len(images) == batch_size:
yield images, image_fns, score_maps, geo_maps, training_masks
images = []
image_fns = []
score_maps = []
geo_maps = []
training_masks = []
except Exception as e:
import traceback
traceback.print_exc()
continue
