任务要求:利用torchvision中的预训练CNN模型来对真实的图像进行分类,预测每张图片的top5类别。
数据: real_image, class_index.json
import torch
from torchvision import models, datasets, transforms
from torch.utils.data import DataLoader, Dataset
from PIL import Image
import numpy as np
import torchvision
import os
import json
import time
import matplotlib.pyplot as plt
%matplotlib inline
构建类别索引词典
f = open('./data/class_index.json')
class_index = json.load(f)
print('class num:', len(class_index))
class_dict = {
int(k): v[1] for k, v in class_index.items()}
print(class_dict)
加载预训练CNN模型
import ssl
alexnet = models.alexnet(pretrained=True)
# import ssl
# resnet = models.resnet50(pretrained=True)
图像缩放、裁剪、转Tensor、归一化
image_transforms = transforms.Compose([
transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(
mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]
)
])
构建测试数据集,迭代返回预处理后的Tensor格式图像和原始图像
class TestDataset():
def __init__(self, root, transforms=None):
imgs = os.listdir(root)
self.imgs = [os.path.join(root, img) for img in imgs]
self.transforms = transforms
def __getitem__(self, index):
img_path = self.imgs[index]
img_pil = Image.open(img_path)
label = None
img_np = np.asarray(img_pil)
data = self.transforms(img_pil)
return data, img_np
def __len__(self):
return len(self.imgs)
test_dir = './data/real_image/'
test_dataset = TestDataset(test_dir, image_transforms)
print('test image num:', test_dataset.__len__())
运行结果如下:
test image num: 20
在测试模式下,对于每张图片显示原始图像,并输出模型预测的top5类别及top1类别
alexnet.eval()
for data, img_np in test_dataset:
img = torch.unsqueeze(data, 0)
output = alexnet(img)
_, index = torch.max(output, 1)
index=index.numpy()
percentage = torch.nn.functional.softmax(output, dim=1)[0] * 100
plt.imshow(img_np, aspect='auto')
plt.show()
print('top1类别:')
print(class_dict[index[0]], percentage[index[0]].item())
_, indices = torch.sort(output, descending=True)
indices=indices.numpy()
print('top5类别:')
for idx in indices[0][:5]:
print((class_dict[idx], percentage[idx].item()))
print()
运行结果如下(部分):