深度学习实践:图像去雨网络实现Pytoch

2023-11-11 15:11

本文主要是介绍深度学习实践:图像去雨网络实现Pytoch,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!

第二集教程链接:http://t.csdn.cn/QxpgD (更详细)

      本文引用 听 风、的博客 图像去雨:超详细手把手写 pytorch 实现代码(带注释)的网络框架,并进行了优化,主要加入了BatchNormalized模块。优化了代码整体框架和书写规范,加入了更多注释。

代码链接:

Kaggle:Derain_Study | Kaggle

Github:Learn_Pytorch/derain-study.ipynb at main · DLee0102/Learn_Pytorch (github.com)

        改进后的代码加入了验证集以观察训练的模型是否过拟合。同时使用了tqdm工具包,方便观察训练进度。在保存模型方面使用了更高效的方法,即保存在验证集上损失最小的模型。

354b4a5cc1ce42178a8b6262567421fd.png

 

 

        数据集采用的是Kaggle上的JRDR - Deraining Dataset的Light数据集,使用了更优化的dataset方法,以使input和label的图片能准确匹配。

4f19311e363f476b9afec88681d7a237.png

 

import os
import torchvision.transforms as transforms
from torch.utils.data import Dataset
from PIL import Image
import torch.optim as optim
import torch
import torch.nn as nn
import torch.nn.functional as F
import matplotlib.pyplot as plt
from torch.utils.data import DataLoader
from tqdm.auto import tqdm
import numpy as np
import re'''
Dataset for Training.
'''
class MyTrainDataset(Dataset):def __init__(self, input_path, label_path):self.input_path = input_pathself.input_files = os.listdir(input_path)self.label_path = label_pathself.label_files = os.listdir(label_path)self.transforms = transforms.Compose([transforms.CenterCrop([64, 64]), transforms.ToTensor(),])def __len__(self):return len(self.input_files)def __getitem__(self, index):label_image_path = os.path.join(self.label_path, self.label_files[index])label_image = Image.open(label_image_path).convert('RGB')'''Ensure input and label are in couple.'''temp = self.label_files[index][:-4]self.input_files[index] = temp + 'x2.png'input_image_path = os.path.join(self.input_path, self.input_files[index])input_image = Image.open(input_image_path).convert('RGB')input = self.transforms(input_image)label = self.transforms(label_image)return input, label'''
Dataset for testing.
'''
class MyValidDataset(Dataset):def __init__(self, input_path, label_path):self.input_path = input_pathself.input_files = os.listdir(input_path)self.label_path = label_pathself.label_files = os.listdir(label_path)self.transforms = transforms.Compose([transforms.CenterCrop([64, 64]), transforms.ToTensor(),])def __len__(self):return len(self.input_files)def __getitem__(self, index):label_image_path = os.path.join(self.label_path, self.label_files[index])label_image = Image.open(label_image_path).convert('RGB')temp = self.label_files[index][:-4]self.input_files[index] = temp + 'x2.png'input_image_path = os.path.join(self.input_path, self.input_files[index])input_image = Image.open(input_image_path).convert('RGB')input = self.transforms(input_image)label = self.transforms(label_image)return input, label
'''
Residual_Network with BatchNormalized.
'''
class Net(nn.Module):def __init__(self):super(Net, self).__init__()self.conv0 = nn.Sequential(nn.Conv2d(6, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU())self.res_conv1 = nn.Sequential(nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU(),nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU())self.res_conv2 = nn.Sequential(nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU(),nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU())self.res_conv3 = nn.Sequential(nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU(),nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU())self.res_conv4 = nn.Sequential(nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU(),nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU())self.res_conv5 = nn.Sequential(nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU(),nn.Conv2d(32, 32, 3, 1, 1),nn.BatchNorm2d(32),nn.ReLU())self.conv = nn.Sequential(nn.Conv2d(32, 3, 3, 1, 1),)def forward(self, input):x = inputfor i in range(6):  # Won't change the number of parameters'''Different from Classification.'''x = torch.cat((input, x), 1)x = self.conv0(x)x = F.relu(self.res_conv1(x) + x)x = F.relu(self.res_conv2(x) + x)x = F.relu(self.res_conv3(x) + x)x = F.relu(self.res_conv4(x) + x)x = F.relu(self.res_conv5(x) + x)x = self.conv(x)x = x + inputreturn x
'''
Check the number of GPU.
'''
print("Let's use", torch.cuda.device_count(), "GPUs!")
'''
Path of Dataset.
'''
input_path = "../input/jrdr-deraining-dataset/JRDR/rain_data_train_Light/rain"
label_path = "../input/jrdr-deraining-dataset/JRDR/rain_data_train_Light/norain"
valid_input_path = '../input/jrdr-deraining-dataset/JRDR/rain_data_test_Light/rain/X2'
valid_label_path = '../input/jrdr-deraining-dataset/JRDR/rain_data_test_Light/norain''''
Check the device.
'''
device = 'cpu'
if torch.cuda.is_available():device = 'cuda''''
Move the Network to the CUDA.
'''
net = Net().to(device)'''
Hyper Parameters.TODO: fine-tuning.
'''
learning_rate = 1e-3
batch_size = 50
epoch = 100
patience = 30
stale = 0
best_valid_loss = 10000'''
Prepare for plt.
'''
Loss_list = []
Valid_Loss_list = []'''
Define optimizer and Loss Function.
'''
optimizer = optim.Adam(net.parameters(), lr=learning_rate)
loss_f = nn.MSELoss()'''
Check the model.
'''
if os.path.exists('./model.pth'): print('Continue train with last model...')net.load_state_dict(torch.load('./model.pth'))
else: print("Restart...")'''
Prepare DataLoaders.Attension:'pin_numbers=True' can accelorate CUDA computing.
'''
dataset_train = MyTrainDataset(input_path, label_path)
dataset_valid = MyValidDataset(valid_input_path, valid_label_path)
train_loader = DataLoader(dataset_train, batch_size=batch_size, shuffle=True, pin_memory=True)
valid_loader = DataLoader(dataset_valid, batch_size=batch_size, shuffle=True, pin_memory=True)'''
START Training ...
'''
for i in range(epoch):
# ---------------Train----------------net.train()train_losses = []'''tqdm is a toolkit for progress bar.'''for batch in tqdm(train_loader):inputs, labels = batchoutputs = net(inputs.to(device))loss = loss_f(outputs, labels.to(device))optimizer.zero_grad()loss.backward()'''Avoid grad to be too BIG.'''grad_norm = nn.utils.clip_grad_norm_(net.parameters(), max_norm=10)optimizer.step()'''Attension:We need set 'loss.item()' to turn Tensor into Numpy, or plt will not work.'''train_losses.append(loss.item())train_loss = sum(train_losses)Loss_list.append(train_loss)print(f"[ Train | {i + 1:03d}/{epoch:03d} ] loss = {train_loss:.5f}")# -------------Validation-------------
'''
Validation is a step to ensure training process is working.
You can also exploit Validation to see if your net work is overfitting.Firstly, you should set model.eval(), to ensure parameters not training.
'''net.eval()valid_losses = []for batch in tqdm(valid_loader):inputs, labels = batch'''Cancel gradient decent.'''with torch.no_grad():outputs = net(inputs.to(device))loss = loss_f(outputs, labels.to(device))valid_losses.append(loss.item())valid_loss = sum(valid_losses)Valid_Loss_list.append(valid_loss)print(f"[ Valid | {i + 1:03d}/{epoch:03d} ] loss = {valid_loss:.5f}")'''Update Logs and save the best model.Patience is also checked.'''if valid_loss < best_valid_loss:print(f"[ Valid | {i + 1:03d}/{epoch:03d} ] loss = {valid_loss:.5f} -> best")else:print(f"[ Valid | {i + 1:03d}/{epoch:03d} ] loss = {valid_loss:.5f}")if valid_loss < best_valid_loss:print(f'Best model found at epoch {i+1}, saving model')torch.save(net.state_dict(), f'model_best.ckpt')best_valid_loss = valid_lossstale = 0else:stale += 1if stale > patience:print(f'No improvement {patience} consecutive epochs, early stopping.')break'''
Use plt to draw Loss curves.
'''
plt.figure(dpi=500)
x = range(epoch)
y = Loss_list
plt.plot(x, y, 'ro-', label='Train Loss')
plt.plot(range(epoch), Valid_Loss_list, 'bs-', label='Valid Loss')
plt.ylabel('Loss')
plt.xlabel('epochs')
plt.legend()
plt.show()

训练结果如下:(显示效果不太好)

ad7d16c58efd4d5d96ab18fb8bb35693.png

 test上实际去雨效果:

原图:

2d801216f83f43a398a996a7f6f2f537.png

 未加入BatchNormalize的效果:

4eddbb740144479dba93683c4f51bb87.jpeg

 加入BatchNormalize后的结果:

5e846e3a16c243a38c07aa3b14e49e9e.jpeg

 可以看到,同样训练论述的情况下,加入BatchNormalize后雨线数目明显减少

 

第二集教程链接:http://t.csdn.cn/QxpgD (更详细)

 

这篇关于深度学习实践:图像去雨网络实现Pytoch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/390938

相关文章

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

关于集合与数组转换实现方法

《关于集合与数组转换实现方法》:本文主要介绍关于集合与数组转换实现方法,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的地方,望不吝赐教... 目录1、Arrays.asList()1.1、方法作用1.2、内部实现1.3、修改元素的影响1.4、注意事项2、list.toArray()2.1、方

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

Java中的雪花算法Snowflake解析与实践技巧

《Java中的雪花算法Snowflake解析与实践技巧》本文解析了雪花算法的原理、Java实现及生产实践,涵盖ID结构、位运算技巧、时钟回拨处理、WorkerId分配等关键点,并探讨了百度UidGen... 目录一、雪花算法核心原理1.1 算法起源1.2 ID结构详解1.3 核心特性二、Java实现解析2.

使用Python实现可恢复式多线程下载器

《使用Python实现可恢复式多线程下载器》在数字时代,大文件下载已成为日常操作,本文将手把手教你用Python打造专业级下载器,实现断点续传,多线程加速,速度限制等功能,感兴趣的小伙伴可以了解下... 目录一、智能续传:从崩溃边缘抢救进度二、多线程加速:榨干网络带宽三、速度控制:做网络的好邻居四、终端交互

java实现docker镜像上传到harbor仓库的方式

《java实现docker镜像上传到harbor仓库的方式》:本文主要介绍java实现docker镜像上传到harbor仓库的方式,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的地... 目录1. 前 言2. 编写工具类2.1 引入依赖包2.2 使用当前服务器的docker环境推送镜像2.2

C++20管道运算符的实现示例

《C++20管道运算符的实现示例》本文简要介绍C++20管道运算符的使用与实现,文中通过示例代码介绍的非常详细,对大家的学习或者工作具有一定的参考学习价值,需要的朋友们下面随着小编来一起学习学习吧... 目录标准库的管道运算符使用自己实现类似的管道运算符我们不打算介绍太多,因为它实际属于c++20最为重要的

Java easyExcel实现导入多sheet的Excel

《JavaeasyExcel实现导入多sheet的Excel》这篇文章主要为大家详细介绍了如何使用JavaeasyExcel实现导入多sheet的Excel,文中的示例代码讲解详细,感兴趣的小伙伴可... 目录1.官网2.Excel样式3.代码1.官网easyExcel官网2.Excel样式3.代码

python实现对数据公钥加密与私钥解密

《python实现对数据公钥加密与私钥解密》这篇文章主要为大家详细介绍了如何使用python实现对数据公钥加密与私钥解密,文中的示例代码讲解详细,感兴趣的小伙伴可以跟随小编一起学习一下... 目录公钥私钥的生成使用公钥加密使用私钥解密公钥私钥的生成这一部分,使用python生成公钥与私钥,然后保存在两个文

MySQL 中 ROW_NUMBER() 函数最佳实践

《MySQL中ROW_NUMBER()函数最佳实践》MySQL中ROW_NUMBER()函数,作为窗口函数为每行分配唯一连续序号,区别于RANK()和DENSE_RANK(),特别适合分页、去重... 目录mysql 中 ROW_NUMBER() 函数详解一、基础语法二、核心特点三、典型应用场景1. 数据分