import torch
import torchvision
from PIL import Image
from torch import nn
image_path = "D:/test pytorch/images/dog.png"
# 读取图片
image = Image.open(image_path)
image = image.convert('RGB')
# png格式的图片是4通道类型,RGB只有三通道,所以调用image = image.convert('RGB')保留其颜色通道。若本来就是三通道,加上这一步也可以。均适用
print(image)
# 网络要求32*32,但是图片显示201*176,对其resize
transform = torchvision.transforms.Compose([torchvision.transforms.Resize((32,32)),
torchvision.transforms.ToTensor()])
image = transform(image)
print(image.shape)
# 加载网络模型
class SUN(nn.Module):
def __init__(self):
super(SUN, 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.Linear(1024, 64),
nn.Linear(64, 10)
)
def forward(self, x):
x =self.model(x)
return x
model = torch.load("sun_0.path")
# 实验结果可看,预测结果不准确,实验结果预测是第二个,但实际在模型中,是第五个类别才是狗,原因是,训练次数的不足
# model = torch.load("sun_29.path")
# 报错显示,在GPU上运行的,使用在CPU上,需要加语句;
# model = torch.load("sun_29.path",map_location = torch.device('cpu'))
print(model)
# 报错:Given groups=1, weight of size [32, 3, 5, 5], expected input[1, 4, 32, 32] to have 3 channels, but got 4 channels instead
image = torch.reshape(image,(1,3,32,32))
# 当模型出现dropout或者batchnormal下述2条就需要下述两条语句
model.eval()
with torch.no_grad():
output = model(image)
output = model(image)
print(output)
print(output.argmax(1))