PyTorch - torchvision.transforms.Compose 几个图像变换组合在一起

PyTorch - torchvision.transforms.Compose 几个图像变换组合在一起
flyfish

transforms是常见的图像变换

torchvision.transforms.Compose是将几个变换组合在一起
原始图片

经过变换的图片

代码如下

import numpy as np
import cv2
import os
import torch
import math
import torchvision.transforms as transforms
from PIL import Image

path="original.jpg"
# 112* 112*3 ，通道顺序B G R，0~255，格式为(H,W,C)
img_original = cv2.imread(path)
img_original = cv2.resize(img_original, (200, 200), interpolation=cv2.INTER_CUBIC)
print(img_original.shape)
for i in range(100):
    transform = transforms.Compose([
        transforms.RandomHorizontalFlip(p=0.5),  # 水平翻转
        transforms.RandomGrayscale(p=0.5),  # 随机灰度
        transforms.ColorJitter(brightness=0.5, contrast=0.5, saturation=0.5, hue=0.5),  # 亮度，对比度，饱和度，色相
    ])
    #CV格式转PIL
    img = Image.fromarray(cv2.cvtColor(img_original, cv2.COLOR_BGR2RGB))
    img = transform(img)
    #PIL转CV格式
    img = cv2.cvtColor(np.asarray(img),cv2.COLOR_RGB2BGR)
    cv2.imwrite(str(i)+'.jpg',img)

在train的阶段和test阶段执行不同的transforms

代码如下

import numpy as np
import cv2
import os
import torch
import math
import torchvision.transforms as transforms
from PIL import Image
path="original.jpg"
# 112* 112*3 ，通道顺序B G R，0~255，格式为(H,W,C)
img_original = cv2.imread(path)
img_original = cv2.resize(img_original, (200, 200), interpolation=cv2.INTER_CUBIC)
transform_dict = {
    'train': transforms.Compose(
        [transforms.Resize(256),

         ]),
    'test': transforms.Compose(
        [transforms.Resize(128),
         ])}

img_original = Image.fromarray(cv2.cvtColor(img_original, cv2.COLOR_BGR2RGB))

phase="train"
transform=transform_dict[phase]
img = transform(img_original)
img = cv2.cvtColor(np.asarray(img),cv2.COLOR_RGB2BGR)
#最终输出文件大小是256 × 256
cv2.imwrite("train.jpg", img)

phase="test"
transform=transform_dict[phase]
img = transform(img_original)
img = cv2.cvtColor(np.asarray(img), cv2.COLOR_RGB2BGR)
#最终输出文件大小是128 × 128
cv2.imwrite("test.jpg", img)