# -*- coding:utf-8 -*-"""数据增强
1. 翻转变换 flip
2. 随机修剪 random crop
3. 色彩抖动 color jittering
4. 平移变换 shift
5. 尺度变换 scale
6. 对比度变换 contrast
7. 噪声扰动 noise
8. 旋转变换/反射变换 Rotation/reflection
"""from PIL import Image, ImageEnhance, ImageOps, ImageFile
import numpy as np
import random
import threading, os, time
import logging
import math
import shutillogger = logging.getLogger(__name__)
ImageFile.LOAD_TRUNCATED_IMAGES = Trueclass DataAugmentation:
"""
包含数据增强的八种方式
"""def __init__(self):
pass@staticmethod
def openImage(image):
return Image.open(image, mode="r")@staticmethod
def randomFlip(image, mode=Image.FLIP_LEFT_RIGHT):
"""
对图像进行上下左右四个方面的随机翻转
:param image: PIL的图像image
:param model: 水平或者垂直方向的随机翻转模式,默认右向翻转
:return: 翻转之后的图像
"""
#random_model = np.random.randint(0, 2)
#flip_model = [Image.FLIP_LEFT_RIGHT, Image.FLIP_TOP_BOTTOM]
#return image.transpose(flip_model[random_model])
return image.transpose(mode)@staticmethod
def randomShift(image):
#def randomShift(image, xoffset, yoffset=None):
"""
对图像进行平移操作
:param image: PIL的图像image
:param xoffset: x方向向右平移
:param yoffset: y方向向下平移
:return: 翻转之后的图像
"""
random_xoffset = np.random.randint(0, math.ceil(image.size[0]*0.2))
random_yoffset = np.random.randint(0, math.ceil(image.size[1]*0.2))
#return image.offset(xoffset = random_xoffset, yoffset = random_yoffset)
return image.offset(random_xoffset)@staticmethod
def randomRotation(image, mode=Image.BICUBIC):
"""
对图像进行随机任意角度(0~360度)旋转
:param mode 邻近插值,双线性插值,双三次B样条插值(default)
:param image PIL的图像image
:return: 旋转转之后的图像
"""
random_angle = np.random.randint(1, 360)
return image.rotate(random_angle, mode)@staticmethod
def randomCrop(image):
"""
对图像随意剪切,裁剪图像大小宽和高的2/3
:param image: PIL的图像image
:return: 剪切之后的图像"""
image_width = image.size[0]
image_height = image.size[1]
crop_image_width = math.ceil(image_width*2/3)
crop_image_height = math.ceil(image_height*2/3)
x = np.random.randint(0, image_width - crop_image_width)
y = np.random.randint(0, image_height - crop_image_height)
random_region = (x, y, x + crop_image_width, y + crop_image_height)
return image.crop(random_region)@staticmethod
def randomColor(image):
"""
对图像进行颜色抖动
:param image: PIL的图像image
:return: 有颜色色差的图像image
"""
random_factor = np.random.randint(0, 31) / 10.# 随机因子
color_image = ImageEnhance.Color(image).enhance(random_factor)# 调整图像的饱和度
random_factor = np.random.randint(10, 21) / 10.# 随机因子
brightness_image = ImageEnhance.Brightness(color_image).enhance(random_factor)# 调整图像的亮度
random_factor = np.random.randint(10, 21) / 10.# 随机因1子
contrast_image = ImageEnhance.Contrast(brightness_image).enhance(random_factor)# 调整图像对比度
random_factor = np.random.randint(0, 31) / 10.# 随机因子
return ImageEnhance.Sharpness(contrast_image).enhance(random_factor)# 调整图像锐度@staticmethod
def randomGaussian(image, mean=0.2, sigma=0.3):
"""
对图像进行高斯噪声处理
:param image:
:return:
"""def gaussianNoisy(im, mean=0.2, sigma=0.3):
"""
对图像做高斯噪音处理
:param im: 单通道图像
:param mean: 偏移量
:param sigma: 标准差
:return:
"""
for _i in range(len(im)):
im[_i] += random.gauss(mean, sigma)
return im# 将图像转化成数组
img = np.asarray(image)
img.flags.writeable = True# 将数组改为读写模式
width, height = img.shape[:2]
try:
img_r = gaussianNoisy(img[:, :, 0].flatten(), mean, sigma)
img_g = gaussianNoisy(img[:, :, 1].flatten(), mean, sigma)
img_b = gaussianNoisy(img[:, :, 2].flatten(), mean, sigma)
img[:, :, 0] = img_r.reshape([width, height])
img[:, :, 1] = img_g.reshape([width, height])
img[:, :, 2] = img_b.reshape([width, height])
except:
img = img
return Image.fromarray(np.uint8(img))@staticmethod
def saveImage(image, path):
try:
image.save(path)
except:
print('not save img: ', path)
passfiles = []
def get_files(dir_path):
global files
if os.path.exists(dir_path):
parents = os.listdir(dir_path)
for parent in parents:
child = os.path.join(dir_path, parent)
if os.path.exists(child) and os.path.isfile(child):
#child = child.split('/')[4:]
#str_child = '/'.join(child)
files.append(child)
elif os.path.isdir(child):
get_files(child)
return files
else:
return Noneif __name__ == '__main__':
times = 2#重复次数
imgs_dir = '/opt/sda/imgData20190322/train'
new_imgs_dir = '/opt/sda/imgData20190322/train_data_augment'
#if os.path.exists(new_imgs_dir):
#shutil.rmtree(new_imgs_dir)
funcMap = {"flip": DataAugmentation.randomFlip,
"rotation": DataAugmentation.randomRotation,
"crop": DataAugmentation.randomCrop,
"color": DataAugmentation.randomColor,
"gaussian": DataAugmentation.randomGaussian
}
#funcLists = {"flip", "rotation", "crop", "color", "gaussian"}
funcLists = {"flip", "rotation", "crop", "gaussian"}global _index
imgs_list = get_files(imgs_dir)
for index_img, img in enumerate(imgs_list):
if index_img != 0 and index_img % 50 == 0:
print('now is dealing %d image' % (index_img) )
tmp_img_dir_list = img.split('/')[:-1]
tmp_img_dir_list[0:len(new_imgs_dir.split('/'))] = new_imgs_dir.split('/')
new_img_dir = '/'.join(tmp_img_dir_list)if not os.path.exists(new_img_dir):
os.makedirs(new_img_dir)
try:
shutil.copy(img, os.path.join(new_img_dir, img.split('/')[-1]))
except:
passimg_name = img.split('/')[-1].split('.')[0]
postfix = img.split('.')[1]#后缀
if postfix.lower() in ['jpg', 'jpeg', 'png', 'bmp']:
image = DataAugmentation.openImage(img)
_index = 1
for func in funcLists:
if func == 'flip':
flip_model = [Image.FLIP_LEFT_RIGHT, Image.FLIP_TOP_BOTTOM]
for model_index in range(len(flip_model)):
new_image = DataAugmentation.randomFlip(image, flip_model[model_index])
img_path = os.path.join(new_img_dir, img_name + '_' + str(_index) + '.' + postfix)
DataAugmentation.saveImage(new_image, img_path)
_index += 1
elif func == 'gaussian':
new_image = DataAugmentation.randomGaussian(image)
img_path = os.path.join(new_img_dir, img_name + '_' + str(_index) + '.' + postfix)
DataAugmentation.saveImage(new_image, img_path)
_index += 1
else:
for _i in range(0, times, 1):
new_image = funcMap[func](image)
img_path = os.path.join(new_img_dir, img_name + '_' + str(_index) + '.' + postfix)
DataAugmentation.saveImage(new_image, img_path)
_index += 1【python|数据增强操作（旋转、翻转、裁剪、色彩变化、高斯噪声等）】
推荐阅读
- YOLO|YOLOX网络结构
- 深度学习|r3det 配环境避雷指南（pytorch版）
- 论文阅读|R3Det: Refined Single-Stage Detector with Feature Refinementfor Rotating Object论文学习
- OCR|使用挤压、哈哈镜、扭曲进行文字图像增强
- 【扫盲】R3Det旋转目标检测训练(win10)
- 【扫盲】R3Det旋转目标检测训练
- 深度学习|pytorch计算分类验证精度acc1，acc5代码
- 深度学习|paddle.nn.functional.cross_entropy中的soft_label时间消耗问题
- 深度学习|使用tensorboard时踩的坑