图像语义分割中的像素精确度问题

from PIL import Image
import numpy as np

def load_labels(image_path):
    # 从标注文件中加载像素级标签
    label_path = image_path.replace('.jpg', '.png')
    label = Image.open(label_path)
    label = np.array(label)     # 转换为numpy数组
    return label

def evaluate_pixel_accuracy(pred_label, gt_label):
    # 计算像素级精确度
    num_correct = np.sum(pred_label == gt_label)
    num_total = pred_label.size
    accuracy = num_correct / num_total
    return accuracy

# 加载预测结果和ground truth
pred_label = load_labels('pred_image.jpg')
gt_label = load_labels('gt_image.jpg')

accuracy = evaluate_pixel_accuracy(pred_label, gt_label)
print("Pixel Accuracy: ", accuracy)

import torch
import torchvision.models as models

# 加载预训练的分割模型
model = models.segmentation.deeplabv3_resnet50(pretrained=True)

# 加载图像数据
image = Image.open('image.jpg')

# 对图像进行预处理
preprocess = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
input_tensor = preprocess(image)
input_batch = input_tensor.unsqueeze(0)

# 使用模型进行预测
with torch.no_grad():
    output = model(input_batch)['out'][0]
pred_label = output.argmax(0).numpy()

# 计算像素级精确度
accuracy = evaluate_pixel_accuracy(pred_label, gt_label)
print("Pixel Accuracy: ", accuracy)