注意力机制打印cam

import cv2
import numpy as np
from PIL import Image

# 加载示例图像
image_path = 'path/to/your/image.jpg'
image = Image.open(image_path).convert('RGB')

# 预处理图像，以符合模型的输入要求
preprocess = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

input_data = preprocess(image).unsqueeze(0)  # 添加 batch 维度
def get_cam(feature_map, spatial_att):
    # 获取权重
    weights = spatial_att.view(spatial_att.size(0), -1).unsqueeze(-1).unsqueeze(-1)
    
    # 将权重乘以特征图
    cam = (weights * feature_map).sum(dim=1, keepdim=True)
    
    # 使用ReLU函数保留正值部分
    cam = F.relu(cam)
    
    # 将CAM调整到0到1之间
    cam = cam - cam.min()
    cam = cam / cam.max()
    
    # 将CAM调整到图像原始尺寸
    cam = F.interpolate(cam, (224, 224), mode='bilinear', align_corners=False)
    
    # 转换为numpy数组
    cam = cam.detach().cpu().numpy()[0, 0]
    
    return cam

# 获取模型输出和对应的中间特征图
output, feature_maps = model(input_data)

# 获取CAM
cam = get_cam(feature_maps[0], feature_maps[1])

# 将CAM叠加在原始图像上
original_image = input_data.detach().cpu().numpy()[0].transpose((1, 2, 0))
cam_heatmap = cv2.applyColorMap(np.uint8(255 * cam), cv2.COLORMAP_JET)
cam_on_image = np.float32(cam_heatmap) / 255 + original_image
cam_on_image = cam_on_image / np.max(cam_on_image)

# 显示CAM叠加的图像
plt.imshow(cam_on_image)
plt.axis('off')
plt.show()