解决PIL和OpenCV读取显示结果不一致问题

现在我们要读取这张图片

使用PIL读取:

img_PIL = Image.open(img_path)
tensor_trans = transforms.ToTensor()
tensor_img = tensor_trans(img_PIL)  # 调用ToTensor()的__call__方法

使用OpenCV读取:

import cv2
img_cv_bgr = cv2.imread(img_path)

这里我用tensorboard工具显示图片,出现以下画面:

这OpenCV读取的咋有点怪，查了查原来是OpenCV读取的通道顺序是BGR，而显示时以及PIL读取的通道顺序是RGB，所以造成OpenCV读取的显示不正常，PIL读取的显示正常的现象。

那咱们调一调OpenCV读取的图像的通道顺序：

img_cv_bgr = cv2.imread(img_path)
# 调整通道顺序
b, g, r = cv2.split(img_cv_bgr)
img_cv_rgb = cv2.merge([r, g, b])

显示结果一致:

以上所有代码:

from PIL import Image
from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms

writer = SummaryWriter('logs')

img_path = r'data\train\ants\0013035.jpg'

img_PIL = Image.open(img_path)
tensor_trans = transforms.ToTensor()
tensor_img = tensor_trans(img_PIL)  # 调用ToTensor()的__call__方法
print(type(tensor_img))
print(tensor_img.shape)
writer.add_image('PIL', tensor_img, 0)

import cv2
img_cv_bgr = cv2.imread(img_path)
print(type(img_cv_bgr))
print(img_cv_bgr.shape)
b, g, r = cv2.split(img_cv_bgr)
img_cv_rgb = cv2.merge([r, g, b])
writer.add_image('openCV', img_cv_bgr, 0, dataformats='HWC')

writer.close()

Out:

<class 'torch.Tensor'>
torch.Size([3, 512, 768])
<class 'numpy.ndarray'>
(512, 768, 3)