体验1.2.0版本自动数据增强

转载地址：https://bbs.huaweicloud.com/forum/thread-128287-1-1.html

作者：御坂8080号

硬件平台：X64

操作系统：Win10 64位

安装截图：

体验教程：自动数据增强

教程地址：自动数据增强

在数据增强的步骤中，比较重要的几步分别是：

定义MindSpore算子到AutoAugment算子的映射

PARAMETER_MAX = 10

def float_parameter(level, maxval):
    return float(level) * maxval /  PARAMETER_MAX

def int_parameter(level, maxval):
    return int(level * maxval / PARAMETER_MAX)

def shear_x(level):
    v = float_parameter(level, 0.3)
    return c_transforms.RandomChoice([c_vision.RandomAffine(degrees=0, shear=(-v,-v)), c_vision.RandomAffine(degrees=0, shear=(v, v))])

def shear_y(level):
    v = float_parameter(level, 0.3)
    return c_transforms.RandomChoice([c_vision.RandomAffine(degrees=0, shear=(0, 0, -v,-v)), c_vision.RandomAffine(degrees=0, shear=(0, 0, v, v))])

def translate_x(level):
    v = float_parameter(level, 150 / 331)
    return c_transforms.RandomChoice([c_vision.RandomAffine(degrees=0, translate=(-v,-v)), c_vision.RandomAffine(degrees=0, translate=(v, v))])

def translate_y(level):
    v = float_parameter(level, 150 / 331)
    return c_transforms.RandomChoice([c_vision.RandomAffine(degrees=0, translate=(0, 0, -v,-v)), c_vision.RandomAffine(degrees=0, translate=(0, 0, v, v))])

def color_impl(level):
    v = float_parameter(level, 1.8) + 0.1
    return c_vision.RandomColor(degrees=(v, v))

def rotate_impl(level):
    v = int_parameter(level, 30)
    return c_transforms.RandomChoice([c_vision.RandomRotation(degrees=(-v, -v)), c_vision.RandomRotation(degrees=(v, v))])

def solarize_impl(level):
    level = int_parameter(level, 256)
    v = 256 - level
    return c_vision.RandomSolarize(threshold=(0, v))

def posterize_impl(level):
    level = int_parameter(level, 4)
    v = 4 - level
    return c_vision.RandomPosterize(bits=(v, v))

def contrast_impl(level):
    v = float_parameter(level, 1.8) + 0.1
    return c_vision.RandomColorAdjust(contrast=(v, v))

def autocontrast_impl(level):
    return c_vision.AutoContrast()

def sharpness_impl(level):
    v = float_parameter(level, 1.8) + 0.1
    return c_vision.RandomSharpness(degrees=(v, v))

def brightness_impl(level):
    v = float_parameter(level, 1.8) + 0.1
    return c_vision.RandomColorAdjust(brightness=(v, v))

定义ImageNet数据集的AutoAugment策略

imagenet_policy = [
    [(posterize_impl(8), 0.4), (rotate_impl(9), 0.6)],
    [(solarize_impl(5), 0.6), (autocontrast_impl(5), 0.6)],
    [(c_vision.Equalize(), 0.8), (c_vision.Equalize(), 0.6)],
    [(posterize_impl(7), 0.6), (posterize_impl(6), 0.6)],
    [(c_vision.Equalize(), 0.4), (solarize_impl(4), 0.2)],

    [(c_vision.Equalize(), 0.4), (rotate_impl(8), 0.8)],
    [(solarize_impl(3), 0.6), (c_vision.Equalize(), 0.6)],
    [(posterize_impl(5), 0.8), (c_vision.Equalize(), 1.0)],
    [(rotate_impl(3), 0.2), (solarize_impl(8), 0.6)],
    [(c_vision.Equalize(), 0.6), (posterize_impl(6), 0.4)],

    [(rotate_impl(8), 0.8), (color_impl(0), 0.4)],
    [(rotate_impl(9), 0.4), (c_vision.Equalize(), 0.6)],
    [(c_vision.Equalize(), 0.0), (c_vision.Equalize(), 0.8)],
    [(c_vision.Invert(), 0.6), (c_vision.Equalize(), 1.0)],
    [(color_impl(4), 0.6), (contrast_impl(8), 1.0)],

    [(rotate_impl(8), 0.8), (color_impl(2), 1.0)],
    [(color_impl(8), 0.8), (solarize_impl(7), 0.8)],
    [(sharpness_impl(7), 0.4), (c_vision.Invert(), 0.6)],
    [(shear_x(5), 0.6), (c_vision.Equalize(), 1.0)],
    [(color_impl(0), 0.4), (c_vision.Equalize(), 0.6)],

    [(c_vision.Equalize(), 0.4), (solarize_impl(4), 0.2)],
    [(solarize_impl(5), 0.6), (autocontrast_impl(5), 0.6)],
    [(c_vision.Invert(), 0.6), (c_vision.Equalize(), 1.0)],
    [(color_impl(4), 0.6), (contrast_impl(8), 1.0)],
    [(c_vision.Equalize(), 0.8), (c_vision.Equalize(), 0.6)],
]

以及在数据读取过程中插入AutoAugment变换

    if do_train:
        dataset = dataset.map(operations=c_vision.RandomSelectSubpolicy(imagenet_policy), input_columns=["image"])

对比体验

最近在复盘一些kaggle的代码，就正好看了一下常用的数据增强手段在不同框架下的实现

使用pytorch的预处理

import torchvision
from torchvision import transforms as T

trans = T.Compose([
    T.Resize(IMAGE_SIZE, IMAGE_SIZE),
    T.RandomHorizontalFlip(p=0.5)
    T.RandomVerticalFlip(0.5),
    T.RandomRotation(90)
])

这里面主要用到了torchvision这个库

其中的API参考是：

https://pytorch.org/vision/stable/transforms.html

而使用mindspore实现相同的功能：

import mindspore.dataset.vision.c_transforms as c_vision
import mindspore.dataset.transforms.c_transforms as c_transforms

trans = [
    c_vision.Resize(IMAGE_SIZE, IMAGE_SIZE),
    c_vision.RandomHorizontalFlip(0.5),
    c_vision.RandomVerticalFlip(0.5),
    c_vision.RandomRotation(degrees=90, resample=Inter.NEAREST, expand=True),
    
]

可以看出来，不需要使用Compose这个API了，相对简洁了一些

API参考：https://www.mindspore.cn/doc/api_python/zh-CN/r1.1/mindspore/mindspore.dataset.vision.html

而在CV的比赛中，我们还会用一个基于OpenCV的库Augmentations，反正就是很快很好用的库

import albumentations as A
trfm = A.Compose([
    A.Resize(IMAGE_SIZE, IMAGE_SIZE),
    A.HorizontalFlip(p=0.5),
    A.VerticalFlip(p=0.6),
    A.RandomRotate90(),
])

API参考：
https://albumentations.readthedocs.io/en/latest/api/augmentations.html#transforms

从这里可以看出来，实现数据增强，用不同框架的方法是差不多的，基础功能上都是差不多的，关键是选择适合的变换方法，才能更好的实现CV比赛的提分

邮箱：hellonexus@qq.com