pytorch进阶————可视化

目录

1、可视化网络结构

1.1 安装torchinfo

2、CNN卷积层可视化

 2.1deach理解

 2.2 CNN特征图可视化

2.3、CNN class activate map 可视化方法

3、利用TensorBoard进行可视化 

3.1 安装 TensorBoard

3.2 TensorBoard可视化基本逻辑

3.3TensorBoard配置与启动

 3.4 Tensorboard图像可视化操作

 3.5 连续变量可视化

 3.6 可视化参数分布

3.7 遇到的问题

---

1、可视化网络结构

#pytorch可视化、
import torch
import torchvision.models as models
from torchvision.models import vgg11
from torchinfo import summary

model = models.vgg11(pretrained= True)
print(dict(model.features.named_children()))

 print(model)，得到的信息只有基础构建的信息，不能显示每一层的shape，为了解决这个问题，引出torchinfo

1.1 安装torchinfo

pip install torchinfo

#pytorch可视化、
#pytorch可视化、
import torch
import torchvision.models as models
from torchvision.models import vgg11
from torchinfo import summary

model = models.vgg11(pretrained= True)
#print(dict(model.features.named_children()))
summary(model,(1,3,224,224))#1代表batch_size,3代表图片通道数，224代表图片的宽高

 

 最终用得到模型每一层的类型、输出shape和数量，模型参数量、模型大小以及一次前向或者反向传播需要的内存大小

2、CNN卷积层可视化

import matplotlib.pyplot as plt
#print(dict(model.features.named_modules()))#1代表batch_size,3代表图片通道数，224代表图片的宽高
conv1 = dict(model.features.named_children())['3']
kernel_set = conv1.weight.detach()#自己理解的意思是把这层第3层网络剥离出来不求其他网络的梯度
num = len(conv1.weight.detach())
print(kernel_set.shape)
for i in range(0,num):
    i_kernel = kernel_set[i]
    plt.figure(figsize = (20,17))
    if len((i_kernel))>1:
        for idx,filter in enumerate(i_kernel):
            plt.axis('off')
            plt.imshow(filter[:,:].detach(),cmap = 'bwr')

 2.1deach理解

import torch
 
a = torch.tensor([1, 2, 3.], requires_grad=True)
print(a.grad)
out = a.sigmoid()
print(out)
 
#添加detach(),c的requires_grad为False
c = out.detach()
print(c)
print(id(out))
print(id(c))
 
#这时候没有对c进行更改，所以并不会影响backward()
out.sum().backward()
print(a.grad)

 按照官方理解,detach()可以使得两个计算图的梯度传递断开，可以看出来返回一个新的tensor，新的tensor和原来的tensor共享数据内存，但不涉及梯度计算，即requires_grad=False。修改其中一个tensor的值，另一个也会改变，因为是共享同一块内存，但如果对其中一个tensor执行某些内置操作，则会报错，例如resize_、resize_as_、set_、transpose_。

 2.2 CNN特征图可视化

与卷积核相对应，输入的原始图像经过每次卷积得到的数据成为特征图，可视化卷积是为了看模型提取那些特征，可视化特征图则是为了看到模型提取到的特征什么样子

import torch
import matplotlib.pyplot as plt
from torchvision import transforms
from torchvision.models import vgg11
from PIL import Image
import numpy as np
                


class Hook(object):
    def __init__(self):
        self.module_name = []
        self.features_in_hook = []
        self.features_out_hook =[]
        
    def __call__(self,module,fea_in,fea_out):#__call__直接将（）应用到自身并执行
        print('hooker working',self)
        self.module_name.append(module.__class__)
        self.features_in_hook.append(fea_in)
        self.features_out_hook.append(fea_out)
        return None
    
    def plot_feature(model,idx,inputs):
        hh = Hook#调用类
        model.features[idx].register_forward_hook(hh)#注册
        
        model.eval()
        __ = model(inputs)
        #forward_model(model,False)#模型前向传播
        print(hh.module_name)
        print((hh.features_in_hook[0][0].shape))
        print((hh.features_out_hook[0].shape))
        out1 = hh.features_out_hook[0]
        total_ft = out1.shape[1]
        first_item = out1[0].cpu().clone()
        plt.figure(figsize = (20,17))
        for ftidx in range(total_ft):
            if fiidx >99:
                break
            ft = first_item[ftidx]
            plt.subplot(10,10,ftidx+1)
            plt.axis('off')
            plt.imshow(ft[ :, :].detach())

2.3、CNN class activate map 可视化方法

CAM作用是判断哪些特征对模型来说是重要的，也衍生出Grad-CAM、等诸多变种

import torch
import matplotlib.pyplot as plt

from torchvision.models import vgg11
from PIL import Image
import numpy as np
model = vgg11(pretrained=True)
img_path = 'C:/Users/CHDT/Desktop/cat.jpg'
img = Image.open(img_path).resize((608,608))
rgb_img = np.float32(img)/255#将原始图像转换为float32格式且在0-1之间
plt.imshow(img)

from pytorch_grad_cam import GradCAM,ScoreCAM,GradCAMPlusPlus,AblationCAM,XGradCAM,EigenCAM,FullGrad
from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
from pytorch_grad_cam.utils.image import show_cam_on_image,preprocess_image
img_tensor = preprocess_image(rgb_img,mean = [0.485,0.486,0.406],std=[0.229,0.224,0.225])
target_layers = [model.features[-1]]
# 选取合适的类激活图，但是ScoreCAM和AblationCAM需要batch_size
cam = GradCAM(model=model,target_layers=target_layers)
targets = [ClassifierOutputTarget(200)]   
# 上方preds需要设定，比如ImageNet有1000类，这里可以设为200

grayscale_cam = cam(input_tensor=img_tensor, targets=targets)
grayscale_cam = grayscale_cam[0, :]
cam_img = show_cam_on_image(rgb_img, grayscale_cam, use_rgb=True)
print(type(cam_img))
Image.fromarray(cam_img)

 2.4 使用FlashTorch实现CNN可视化

#使用FlashTorch工具快速实现CNN可视化
import matplotlib.pyplot as plt
from PIL import Image
import torchvision.models as models
from flashtorch.utils import apply_transforms,load_image
from flashtorch.saliency import Backprop
model = models.alexnet(pretrained = True)
backprop = Backprop(model)
img_path = 'C:/Users/CHDT/Desktop/cat.jpg'
img = Image.open(img_path).resize((224,224))

#rgb_img = np.float32(img)/255
owl = apply_transforms(img)
target_class = 24
backprop.visualize(owl,target_class,guided=True,use_gpu = True)

遇到一个问题是这样，用自己本地的图像，出现错误，目前还在研究，未解决 ，继续学习下

2.4.1可视化卷积核

#可视化卷积核
import matplotlib.pyplot as plt
from PIL import Image
import torchvision.models as models
from flashtorch.utils import apply_transforms,load_image
from flashtorch.saliency import Backprop
from flashtorch.activmax import GradientAscent
model = models.vgg16(pretrained= True)
g_ascent = GradientAscent(model.features)
conv5_1 = model.features[24]
conv5_1_filters = [45,271,363,489]
g_ascent.visualize(conv5_1,conv5_1_filters,title ='VGG16:covn5_1')

3、利用TensorBoard进行可视化 

3.1 安装 TensorBoard

pip install tensorboardX

 也可以使用torch自带的。

3.2 TensorBoard可视化基本逻辑

记录包括模型权重、训练损失等等。》》》》》记录下的内容可以已网页形式加以可视化。

3.3TensorBoard配置与启动

#tensorbard
from tensorboardX import SummaryWriter
writer = SummaryWriter('./runs')

上述操作实例化SummaryWriter为变量Writer,指定输出目录下的run目录，，之后类似于模型的权重文件 epoch以及模型损失都可以保存至在run目录。

#tensorbard模型可视化

import torch.nn as nn
import torchvision.models as models
net = models.vgg11()
print(net)

 3.4 Tensorboard图像可视化操作

#tensorboard图像可视化
import torchvision
from torchvision import datasets,transforms
from torch.utils.data import DataLoader
transforms_train = transforms.Compose(
        [transforms.ToTensor()])#利用transforms将图像变换为Tensor形式
transforms_test = transforms.Compose(
        [transforms.ToTensor()])
train_data = datasets.CIFAR10(".",train = True,download =True,transform=transforms_train)
test_data = datasets.CIFAR10(".",train = False,download =True,transform=transforms_test)#划分数据集
train_loader = DataLoader(train_data,batch_size =64,shuffle = True)#打乱数据集一次取64张图像进行训练
test_loader=DataLoader(test_data,batch_size=64)
images,labels = next(iter(train_loader))
#查看一张图片
writer = SummaryWriter('./pytorch_tb')
writer.add_image('images[0]',images[0])
writer.close()
notebook.start("--logdir .//Desktop/pytorch_tb")#启动tensorboard

发现了错误是未启动tensorboard造成的,>>>>>因此添加notebook.start("--logdir .//Desktop/pytorch_tb")#启动tensorboard

#多张图像直接写入
writer = SummaryWriter('./pytorch_tb')
writer.add_images('images',images,global_step=0)
writer.close()

#将多张图片拼接成一整图片
writer = SummaryWriter('./pytorch_xs')
img_grid = torchvision.utils.make_grid(images)
writer.add_image('image_grid',img_grid)
writer.close()

 3.5 连续变量可视化

#连续变量可视化
writer = SummaryWriter('./xiao')
for i in range(500):
    x= i
    y = x*2
    writer.add_scalar("x",x,i)
    writer.add_scalar("y",y,i)
writer.close()
    

 3.6 可视化参数分布

#参数分布可视化
import torch
import numpy as np
def norm(mean,std):
    t = std*torch.randn((100,20))+mean
    return t
writer = SummaryWriter('./Desktop/xiaosun')

for step,mean in enumerate(range(-10,10,1)):
    w = norm(mean,1)
    writer.add_histogram("w",w,step)
    writer.flush()
writer.close()

 

主要实现是通过SummaryWriter,然后通过add_xxx()函数实现

3.7 遇到的问题

 连续打开文件运行tensorboard，运行的总是这一个图像，不发生变化

》》》》》》》》》》》》》》》》》》》》》》》》》》》》

上面程序运行自己本地图片发生错误 未解决 

继续学习

https://zhuanlan.zhihu.com/p/389738863

https://github.com/datawhalechina