【pytorch卷1】=3=卷积提取提取图片轮廓

目录

1. 用mping载入图片

2. 用transforms.ToTensor图片转为张量，并（自动）归一化

3. 定义卷积核

4. 图片卷积函数处理

---

1. 用mping载入图片

import matplotlib.pyplot as plt
import torch
import matplotlib.image as maping
import torchvision.transforms as transforms

#import the underlying picture by apply maping
myimg = maping.imread('C:/Users/liyzc/Desktop/胖胖/和胖胖在宁波/IMG_20201007_122412.jpg')

#show the picture
plt.imshow(myimg)
plt.axis('off')
plt.show()
print(myimg.shape)

2. 用transforms.ToTensor图片转为张量，并（自动）归一化

pil2tensor = transforms.ToTensor()
rgb_image = pil2tensor(myimg)
print(rgb_image)
print(rgb_image[0][0])
print(rgb_image.shape)

3. 定义卷积核（方法参考）

sobelfilter = torch.tensor([[-1., 0, 1.],
                       [-2., 0, 2.],
                       [-1., 0, 1.]]*3).reshape([1,3,3,3])

print(sobelfilter)
#note: *3 means tripling this ARRAY. compare:
# tensor([[-1., 0, 1.],
          [-2., 0, 2.],
          [-1., 0, 1.]])*3

4. 图片卷积函数处理

op = torch.nn.functional.conv2d(rgb_image.unsqueeze(0),soberfilter, stride=3,padding=(1,1))
#tensor转化为图片数据
ret = (op - op.min())/(op.max()-op.min())
ret = (ret.clamp(0.,1.)*255).int()#逆归一化搞回去，并转化乘整数
plt.imshow(ret.squeeze(), cmap='ocean_r')
plt.axis('off')
plt.show()