单张图像经过简单模型的卷积可视化

使用kaggle dog bless 数据集
不使用dataloader的情况下
手动将图像转为tensor，并简单搭建只有一个卷积的神经网络
经过模型卷积后的可视乎效果

知识点：

1. cv2->tensor
2. model 搭建
3. tensor->cv2.show

import cv2
import torch.nn as nn
import torch
import numpy
import torch.nn.functional as F
class LeNet_1(nn.Module):
    def __init__(self):
        super(LeNet_1, self).__init__()
        self.conv1 = nn.Sequential(
            nn.Conv2d(in_channels=3, out_channels=3, kernel_size=3, stride=1,padding=1)
        )
        self.maxpool = nn.Sequential(
            nn.MaxPool2d(kernel_size=2)
        )
        self.fc1 = nn.Sequential(
            nn.Linear(in_features= 512*512*3 , out_features=120)
        )

    def forward(self,x):
        x = self.conv1(x)
        # x = x.view(-1, 512*512*3) #后面有多分类，整理成[1,n]维度
        # x = self.fc1(x) # 全连接层，输出[1,classes]
        # x = self.maxpool(x)#降低一半分辨率
        return x


if __name__ == '__main__':
    model = LeNet_1()
    batchsize =1
    channel = 3
    img_w = 512
    img_h = 512
    impath = "../dogDataset/train/0a0c223352985ec154fd604d7ddceabd.jpg"
    img = cv2.imread(impath)#读入img
    img = cv2.resize(img,(img_w,img_h))#调整长宽
    img_tensor = torch.from_numpy(img)#手动转tensor[512，512,3]
    # tensorx = torch.rand(batchsize, channel, img_w, img_h)
    img_tensor = img_tensor.permute(2,0,1)  # 手动维度转换[3,512,512]
    img_tensor = img_tensor.unsqueeze(0) # 添加维度[1,3,512,512]==>[batchsize, channel, img_w, img_h]
    img_tensor = img_tensor.type(torch.float32)#需要float32采用用CEloss计算
    out = model.forward(img_tensor)
    
    #接下来的代码仅仅为了可视化，不属于模型必须
    img = out.type(torch.uint8).squeeze()#去除第一个维度
    img = img.permute(1,2,0)#手动维度转换[3,512,512]==》[512,512,3]
    img = img.numpy() # tensor -> np
    outmat = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
    cv2.imshow("dfdfd", outmat)  # can't show because it has been transform to tensor
    cv2.waitKey(0)