深度学习框架输出可视化中间层特征与类激活热力图

2023-12-29 13:28

本文主要是介绍深度学习框架输出可视化中间层特征与类激活热力图,希望对大家解决编程问题提供一定的参考价值,需要的开发者们随着小编来一起学习吧!

有时候为了分析深度学习框架的中间层特征,我们需要输出中间层特征进行分析,这里提供一个方法。

(1)输出中间特征层名字

导入所需的库并加载模型

import matplotlib.pyplot as plt
import torch
import torch.nn as nn
from torch.nn import functional as F
from torchvision import transforms
import numpy as np
from PIL import Image
from collections import OrderedDict
import cv2
from models.xxx import Model  # 加载自己的模型, 这里xxx是自己模型名字
import os
device = torch.device('cuda:0')
model = Model().to(device)
print(model)

输出如下,这里我只截取了部分模型中间层输出

Model((res): ResNet50((conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3))(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)(layer1): Sequential((0): ResNet50DownBlock((conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(extra): Sequential((0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)))(1): ResNet50BasicBlock((conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True))(2): ResNet50BasicBlock((conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)))(layer2): Sequential((0): ResNet50DownBlock((conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(extra): Sequential((0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2))(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)))(1): ResNet50BasicBlock((conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True))(2): ResNet50BasicBlock((conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True))(3): ResNet50DownBlock((conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(extra): Sequential((0): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1))(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True))))(layer3): Sequential((0): ResNet50DownBlock((conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(extra): Sequential((0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2))(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)))(1): ResNet50BasicBlock((conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True))(2): ResNet50BasicBlock((conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True))(3): ResNet50DownBlock((conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(extra): Sequential((0): Conv2d(1024, 1024, kernel_size=(1, 1), stride=(1, 1))(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)))(4): ResNet50DownBlock((conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(extra): Sequential((0): Conv2d(1024, 1024, kernel_size=(1, 1), stride=(1, 1))(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)))(5): ResNet50DownBlock((conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1))(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1))(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)(extra): Sequential((0): Conv2d(1024, 1024, kernel_size=(1, 1), stride=(1, 1))(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True))))

(2)加载并处理图像

img_path = './dataset//val_data/images/100_0019_0165-11.jpg'
img = Image.open(img_path)
imgarray = np.array(img)/255.0
# plt.figure(figsize=(8, 8))
# plt.imshow(imgarray)
# plt.axis('off')
# plt.show()

加载后如下

将图片处理成模型可以预测的形式

# 处理图像
transform = transforms.Compose([transforms.Resize([512, 512]),transforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
input_img = transform(img).unsqueeze(0)  # unsqueeze(0)用于升维
# print(input_img.shape)   # torch.Size([1, 3, 512, 512])

(3)可视化中间层

1.定义钩子函数

# 定义钩子函数
activation = {}  # 保存获取的输出
def get_activation(name):def hook(model, input, output):activation[name] = output.detach()return hook

2.可视化中间层特征,这里选择了一个层,其他的自己可以类推

# 可视化中间层特征
checkpoint = torch.load('./checkpoint_best.pth')  # 加载一下权重
model.load_state_dict(checkpoint['model'])
model.eval()
model.res.layer1[2].register_forward_hook(get_activation('bn3'))  #resnet50 layer1中第三个模块的bn3注册钩子
input_img = input_img.to(device)  # cpu数据转一下gpu,这个看你会不会报错,我的不转会报错
_ = model(input_img)
bn3 = activation['bn3']   # 结果将保存在activation字典中  bn3输出<class 'torch.Tensor'>, tensor是无法用plt正常显示的
# print(bn3.shape)  # 调试到这里基本成功了
bn3 = bn3.cpu().numpy() # 转一下numpy,  shape:(1,256, 128, 128) 
plt.figure(figsize=(8,8))
plt.imshow(bn3[0][0], cmap='gray')  # bn3[0][0]  shape:(128, 128)
plt.axis('off')
# # shape:(128, 128)
plt.show()

可视化结果

(4)利用循环输出多张图像可视化中间层

整合上面的代码,利用循环输出验证集中的多张图像中的可视化中间层

# 加载依赖包
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
from torch.nn import functional as F
from torchvision import transforms
import numpy as np
from PIL import Image
from collections import OrderedDict
import cv2
from models.M_SFANet import Model
import os
import glob# 定义钩子函数
activation = {}  # 保存获取的输出
def get_activation(name):def hook(model, input, output):activation[name] = output.detach()return hook# 加载模型
device = torch.device('cuda:0')
model = Model().to(device)checkpoint = torch.load('./checkpoint_best.pth')  # 加载一下权重
model.load_state_dict(checkpoint['model'])
model.eval()
model.res.layer1[2].register_forward_hook(get_activation('bn3'))  #resnet50 layer1中第三个模块的bn3注册钩子,如果需要其他层数就用其他的# 利用循环输出多个可视化中间层#读取需要输出特征的图像
DATA_PATH = f"./val_data/"
img_list = glob.glob(os.path.join(DATA_PATH, "images", "*.jpg"))    # image 路径
img_list.sort()
for idx in range(0, len(img_list)):img_name = img_list[idx].split('/')[-1].split('.')[0]  # 获取文件名img = Image.open(img_list[idx])  # 可以读到图片imgarray = np.array(img)/255.0# 处理图像transform = transforms.Compose([transforms.Resize([512, 512]),transforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])input_img = transform(img).unsqueeze(0)  # unsqueeze(0)用于升维input_img = input_img.to(device)  # cpu数据转一下gpu,这个看你会不会报错,我的会报错_ = model(input_img)bn3 = activation['bn3']   # 结果将保存在activation字典中  bn3输出<class 'torch.Tensor'>, tensor是无法用plt正常显示的bn3 = bn3.cpu().numpy() plt.figure(figsize=(8,8))plt.imshow(bn3[0][0], cmap='jet')  # bn3[0][0]  shape:(128, 128)plt.axis('off')# # shape:(128, 128)plt.savefig('./feature_out/res50/layer1/{}_res50_layer1'.format(img_name), bbox_inches='tight', pad_inches=0.05, dpi=300)

保存至文件夹中如下

---------------------------------------------------更新于2023.1121.28 -----------------------------------------

(5)利用循环输出多张图像类激活热力图

使用类激活热力图,能观察模型对图像识别的关键位置。

这里接着上面的获得的特征图进一步得到类激活热力图

接着上面获取到bn3,代码如下

    bn3 = activation['bn3']   # 结果将保存在activation字典中  bn3输出<class 'torch.Tensor'>, tensor是无法用plt正常显示的'''以下代码用于输出特征图bn3 = bn3.cpu().numpy() plt.figure(figsize=(8,8))plt.imshow(bn3[0][0], cmap='jet')  # bn3[0][0]  shape:(128, 128)plt.axis('off')# # shape:(128, 128)plt.savefig('./feature_out/res50/layer4/{}_res50_layer4'.format(img_name), bbox_inches='tight', pad_inches=0.05, dpi=300)'''# 将特征图用类热力图形式叠加到原图中bn3 = bn3[0][0].cpu().numpy()bn3 = np.maximum(bn3, 0)bn3 /= np.max(bn3)# plt.matshow(bn3)# plt.show()# img1 = cv2.imread('./dataset/ShanghaiTech/part_A_final/val_data/images/100_0019_0165-11.jpg')img1 = cv2.cvtColor(np.asarray(img), cv2.COLOR_RGB2BGR)  # PIL Image转一下cv2bn3 = cv2.resize(bn3, (img1.shape[1], img1.shape[0]))bn3 = np.uint8(255 * bn3)bn3 = cv2.applyColorMap(bn3, cv2.COLORMAP_JET)heat_img = cv2.addWeighted(img1, 1, bn3, 0.5, 0)cv2.imwrite('./heatmap_out/res50/layer1/{}_res50_layer1.jpg'.format(str(img_name)), heat_img)

输出如下

这篇关于深度学习框架输出可视化中间层特征与类激活热力图的文章就介绍到这儿,希望我们推荐的文章对编程师们有所帮助!



http://www.chinasem.cn/article/549569

相关文章

在Linux中改变echo输出颜色的实现方法

《在Linux中改变echo输出颜色的实现方法》在Linux系统的命令行环境下,为了使输出信息更加清晰、突出,便于用户快速识别和区分不同类型的信息,常常需要改变echo命令的输出颜色,所以本文给大家介... 目python录在linux中改变echo输出颜色的方法技术背景实现步骤使用ANSI转义码使用tpu

深度解析Java DTO(最新推荐)

《深度解析JavaDTO(最新推荐)》DTO(DataTransferObject)是一种用于在不同层(如Controller层、Service层)之间传输数据的对象设计模式,其核心目的是封装数据,... 目录一、什么是DTO?DTO的核心特点:二、为什么需要DTO?(对比Entity)三、实际应用场景解析

深度解析Java项目中包和包之间的联系

《深度解析Java项目中包和包之间的联系》文章浏览阅读850次,点赞13次,收藏8次。本文详细介绍了Java分层架构中的几个关键包:DTO、Controller、Service和Mapper。_jav... 目录前言一、各大包1.DTO1.1、DTO的核心用途1.2. DTO与实体类(Entity)的区别1

Spring 框架之Springfox使用详解

《Spring框架之Springfox使用详解》Springfox是Spring框架的API文档工具,集成Swagger规范,自动生成文档并支持多语言/版本,模块化设计便于扩展,但存在版本兼容性、性... 目录核心功能工作原理模块化设计使用示例注意事项优缺点优点缺点总结适用场景建议总结Springfox 是

深度解析Python装饰器常见用法与进阶技巧

《深度解析Python装饰器常见用法与进阶技巧》Python装饰器(Decorator)是提升代码可读性与复用性的强大工具,本文将深入解析Python装饰器的原理,常见用法,进阶技巧与最佳实践,希望可... 目录装饰器的基本原理函数装饰器的常见用法带参数的装饰器类装饰器与方法装饰器装饰器的嵌套与组合进阶技巧

深度解析Spring Boot拦截器Interceptor与过滤器Filter的区别与实战指南

《深度解析SpringBoot拦截器Interceptor与过滤器Filter的区别与实战指南》本文深度解析SpringBoot中拦截器与过滤器的区别,涵盖执行顺序、依赖关系、异常处理等核心差异,并... 目录Spring Boot拦截器(Interceptor)与过滤器(Filter)深度解析:区别、实现

深度解析Spring AOP @Aspect 原理、实战与最佳实践教程

《深度解析SpringAOP@Aspect原理、实战与最佳实践教程》文章系统讲解了SpringAOP核心概念、实现方式及原理,涵盖横切关注点分离、代理机制(JDK/CGLIB)、切入点类型、性能... 目录1. @ASPect 核心概念1.1 AOP 编程范式1.2 @Aspect 关键特性2. 完整代码实

SpringBoot开发中十大常见陷阱深度解析与避坑指南

《SpringBoot开发中十大常见陷阱深度解析与避坑指南》在SpringBoot的开发过程中,即使是经验丰富的开发者也难免会遇到各种棘手的问题,本文将针对SpringBoot开发中十大常见的“坑... 目录引言一、配置总出错?是不是同时用了.properties和.yml?二、换个位置配置就失效?搞清楚加

Python的端到端测试框架SeleniumBase使用解读

《Python的端到端测试框架SeleniumBase使用解读》:本文主要介绍Python的端到端测试框架SeleniumBase使用,具有很好的参考价值,希望对大家有所帮助,如有错误或未考虑完全... 目录SeleniumBase详细介绍及用法指南什么是 SeleniumBase?SeleniumBase

HTML5 中的<button>标签用法和特征

《HTML5中的<button>标签用法和特征》在HTML5中,button标签用于定义一个可点击的按钮,它是创建交互式网页的重要元素之一,本文将深入解析HTML5中的button标签,详细介绍其属... 目录引言<button> 标签的基本用法<button> 标签的属性typevaluedisabled