PIL
- transforms.ToTensor()
from torchvision import transforms
transform_tensor = transforms.Compose([
# PIL.Image/numpy.ndarray -> Tensor, including range [0, 255] -> [0.0,1.0]
transforms.ToTensor(),
)
注意：transforms.ToTensor() 内含归一至 [0.0,1.0] 操作
from PIL import Image
# preproccess
img_path = "snorlax.png"
img = Image.open(img_path).convert('RGB') # 读取图像
img_tensor = transform_tensor(img) # 归一化到 [0.0,1.0]
print(img_tensor.shape) # torch.Size([3, 959, 959])
# convert to PILImage, and show
img_PIL = transforms.ToPILImage()(img_tensor).convert('RGB')
print(img_PIL.size, img_PIL.mode) # (959, 959) RGB
img_PIL.show()
- transforms.Normalize()
from torchvision import transforms
# transforms.ToTensor()
transform_norm = transforms.Compose([
# PIL.Image/numpy.ndarray -> Tensor, including range [0, 255] -> [0.0,1.0]
transforms.ToTensor(),
# range [0.0,1.0] -> [-1.0, 1.0] # channel=（channel-mean）/std
transforms.Normalize(mean=(0.5, 0.5, 0.5), std=(0.5, 0.5, 0.5))
]
)
将 [0.0,1.0] 归一至 [-1.0, 1.0] ，图像色域会出现偏差 ⤵️
- transforms.RandomCrop()
from torchvision import transforms
transform_rc = transforms.Compose([
transforms.ToTensor(),
transforms.ToPILImage(),
transforms.RandomCrop((450, 450)),
]
)
from PIL import Image
img_path = "snorlax.png"
img = Image.open(img_path).convert('RGB')
img_PIL = transform_rc(img)
img_PIL.show()
numpy.ndarray
from torchvision import transforms
import cv2
import numpy as np
img = cv2.imread(img_path) # 读取图像 # img: <class 'numpy.ndarray'> (959, 959, 3) H*W*C
img_tensor = transform_tensor(img) # 归一化到 [0.0,1.0]
print(img_tensor.shape) # torch.Size([3, 959, 959]) C*H*W
# conert to numpy.ndarray, and show
img_arr = img_tensor.numpy()*255
img_arr = img_arr.astype('uint8')
img_arr = np.transpose(img_arr, (1, 2, 0))
print(img_arr.shape) # (959, 959, 3) H*W*C
cv2.imshow('img_arr', img_arr)
cv2.waitKey()
