完整的模型验证套路
完整的模型验证（测试，demo）套路-利用已经训练好的模型，然后给它提供输入
如果模型训练好了，这个程序就是我们可以对外提供的一个实际的应用
图片size=1279x1706，刚才的网络模型只能输入32×32，所以需要进行一下resize torchvision.transforms.Resize
使用cpu训练的10轮验证dog
import torch
import torchvision
from PIL import Image
from torch import nn
image_path = "imgs/dog.jpg"
image = Image.open(image_path)
print(image)
image = image.convert('RGB')
# 讲者图片png格式，是四个通道，除了RGB三通道外，还有一个透明度通道，所以调用image = image.convert('RGB')，保留其颜色通道
# 当然，如果图片本来就是三个颜色通道，经过此操作，不变，加上这一步后，可以适应png，jpg各种格式的图片
transform = torchvision.transforms.Compose([torchvision.transforms.Resize((32, 32)),
torchvision.transforms.ToTensor()])
image = transform(image)
print(image.shape)
# 搭建神经网络
class Pongber(nn.Module):
def __init__(self):
super(Pongber, self).__init__()
self.model = nn.Sequential(
nn.Conv2d(3, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 64, 5, 1, 2),
nn.MaxPool2d(2),
# nn.Flatten(),
nn.Flatten(start_dim=0),
nn.Linear(1024, 64),
nn.Linear(64, 10)
)
def forward(self, x):
x = self.model(x)
return x
model = torch.load("P28_train_cpu/pongber_10.pth", map_location=torch.device('cpu'))
print(model)
image = torch.reshape(image, (1, 3, 32, 32))
model.eval()
with torch.no_grad():
output = model(image)
print(output)
print(output.argmax(1))
返回：
<PIL.PngImagePlugin.PngImageFile image mode=RGBA size=546x580 at 0x2DA9B0422C0>
torch.Size([3, 32, 32])
Pongber(
(model): Sequential(
(0): Conv2d(3, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(2): Conv2d(32, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(6): Flatten(start_dim=1, end_dim=-1)
(7): Linear(in_features=1024, out_features=64, bias=True)
(8): Linear(in_features=64, out_features=10, bias=True)
)
)
tensor([[-1.3282, -2.0507, -1.2791, 2.7573, 0.1886, 1.1080, 0.6323, 1.9815,
-2.5341, 1.2562]])
tensor([3])
验证失败
使用Google Colab上的T4 GPU训练60轮验证dog
import torch
import torchvision
from PIL import Image
from torch import nn
image_path = "imgs/dog.png"
image = Image.open(image_path)
print(image)
image = image.convert('RGB')
# 讲者图片png格式，是四个通道，除了RGB三通道外，还有一个透明度通道，所以调用image = image.convert('RGB')，保留其颜色通道
# 当然，如果图片本来就是三个颜色通道，经过此操作，不变，加上这一步后，可以适应png，jpg各种格式的图片
transform = torchvision.transforms.Compose([torchvision.transforms.Resize((32, 32)),
torchvision.transforms.ToTensor()])
image = transform(image)
print(image.shape)
# 搭建神经网络
class Pongber(nn.Module):
def __init__(self):
super(Pongber, self).__init__()
self.model = nn.Sequential(
nn.Conv2d(3, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 64, 5, 1, 2),
nn.MaxPool2d(2),
# nn.Flatten(),
nn.Flatten(start_dim=0),
nn.Linear(1024, 64),
nn.Linear(64, 10)
)
def forward(self, x):
x = self.model(x)
return x
model = torch.load("P28_train_cpu/pongber_10.pth", map_location=torch.device('cpu'))
# model = torch.load("pongber_60.pth", map_location=torch.device('cpu'))
print(model)
image = torch.reshape(image, (1, 3, 32, 32))
model.eval()
with torch.no_grad():
output = model(image)
print(output)
print(output.argmax(1))
返回：
<PIL.PngImagePlugin.PngImageFile image mode=RGBA size=546x580 at 0x24BBE516260>
torch.Size([3, 32, 32])
Pongber(
(model): Sequential(
(0): Conv2d(3, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(2): Conv2d(32, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(6): Flatten(start_dim=1, end_dim=-1)
(7): Linear(in_features=1024, out_features=64, bias=True)
(8): Linear(in_features=64, out_features=10, bias=True)
)
)
tensor([[ 0.6350, -12.2220, 1.5498, 13.1955, -1.2640, 13.4426, 4.3041,
2.7227, -18.3696, -4.1950]])
tensor([5])
验证成功
验证一下airplane，选择了qian35和su57来验证
import torch
import torchvision
from PIL import Image
from torch import nn
# image_path = "imgs/dog.png"
image_path = "imgs/qian35.png"
# image_path = "imgs/su57.png"
image = Image.open(image_path)
print(image)
image = image.convert('RGB')
# 讲者图片png格式，是四个通道，除了RGB三通道外，还有一个透明度通道，所以调用image = image.convert('RGB')，保留其颜色通道
# 当然，如果图片本来就是三个颜色通道，经过此操作，不变，加上这一步后，可以适应png，jpg各种格式的图片
transform = torchvision.transforms.Compose([torchvision.transforms.Resize((32, 32)),
torchvision.transforms.ToTensor()])
image = transform(image)
print(image.shape)
# 搭建神经网络
class Pongber(nn.Module):
def __init__(self):
super(Pongber, self).__init__()
self.model = nn.Sequential(
nn.Conv2d(3, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 32, 5, 1, 2),
nn.MaxPool2d(2),
nn.Conv2d(32, 64, 5, 1, 2),
nn.MaxPool2d(2),
# nn.Flatten(),
nn.Flatten(start_dim=0),
nn.Linear(1024, 64),
nn.Linear(64, 10)
)
def forward(self, x):
x = self.model(x)
return x
# model = torch.load("P28_train_cpu/pongber_10.pth", map_location=torch.device('cpu'))
model = torch.load("pongber_60.pth", map_location=torch.device('cpu'))
print(model)
image = torch.reshape(image, (1, 3, 32, 32))
model.eval()
with torch.no_grad():
output = model(image)
print(output)
print(output.argmax(1))
qian35返回：
<PIL.PngImagePlugin.PngImageFile image mode=RGBA size=577x423 at 0x1BADC1D22C0>
torch.Size([3, 32, 32])
Pongber(
(model): Sequential(
(0): Conv2d(3, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(2): Conv2d(32, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(6): Flatten(start_dim=1, end_dim=-1)
(7): Linear(in_features=1024, out_features=64, bias=True)
(8): Linear(in_features=64, out_features=10, bias=True)
)
)
tensor([[ 14.9094, -20.2656, 15.4033, 7.3129, 8.6690, 2.1221, -3.3266,
2.6043, -12.5161, -11.7391]])
tensor([2])
验证失败（识别成了鸟，与照片的选取和模型的准确率有关系）
su57返回：
<PIL.PngImagePlugin.PngImageFile image mode=RGBA size=437x612 at 0x1A38AF062C0>
torch.Size([3, 32, 32])
Pongber(
(model): Sequential(
(0): Conv2d(3, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(2): Conv2d(32, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(4): Conv2d(32, 64, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
(5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(6): Flatten(start_dim=1, end_dim=-1)
(7): Linear(in_features=1024, out_features=64, bias=True)
(8): Linear(in_features=64, out_features=10, bias=True)
)
)
tensor([[ 8.8159, -4.9627, -0.9012, -0.9941, -1.0859, 3.4090, -13.1614,
4.1117, 2.1128, 1.1760]])
tensor([0])
验证成功
test就是把训练模型应用到实际环境过程中
选择图片数据
resize符合模型
加载已经训练好的模型，不同环境中训练的模型应用到不同的环境中，需要经过映射
要求输入单个图片的时候，因为网络训练过程中需要batchsize，没有batchsize的话尺度是不符合的，需要reshape
model.eval()最好要写，因为假如我们的网络中正好出现了dropout或者batchnormal时也许预测就会有问题
最后得到预测结果，像分类结果的输出不利于解读，转化成利于解读的方式
原始资料地址：
完整的模型验证套路
如有侵权联系删除 仅供学习交流使用