PyTorchプロジェクトの応用例(三)共通の画像分類モデルによる画像分類(コードと操作方法付)

背景:画像データから画像をロードし、ラベルに基づいてトレーニングする必要があります.簡単にフォルダに画像を直接入れてラベルとトレーニングデータを生成し、モデルをトレーニングします.実画像から訓練されたモデルへの転換.
コードアドレス(直接実行可能):githubアドレス:https://github.com/Xingxiangrui/image_classification_with_pytorch
copy 4のコードを直接使用することもできます.
目次
一、小サンプルの運行とデバッグ
1.1データセット
1.2ラベル形式
1.3最小データセット運転
1.4可能性のあるエラーおよび解決策(見ざるを得ないものはない)
二、大サンプル量データセットの生成
2.1画像フォーマット
2.2ラベル形式
2.3バッチ生成データとラベル
画像名読み出し
シーヶンス乱れ
リストを2つのリストに分割
画像の読み出しと書き込み
ラベルの生成
三、訓練と検証
3.1データのロード
3.2データロード関数
3.3モデルトレーニングと検証
3.4モデルのトレーニングと保存
四、コード
一、小サンプルの運行とデバッグ
まず小サンプルでプログラムを実行し、プログラムが正しく実行されることを保証し、大サンプルで実験する.コードは最後を参照してください.
1.1データセット
コードにデータセットをロードする方法は次のとおりです.

    print("Load dataset......")
    image_datasets = {x: customData(img_path='data/',
                                    txt_path=('data/TxtFile/' + x + '.txt'),
                                    data_transforms=data_transforms,
                                    dataset=x) for x in ['train', 'val']}

1.2ラベル形式
場所data/Txtfile/フォルダの下
2つのファイルtxtとval.txt
フォルダ内の各行は、*****121のようなファイルのパスである.jpg(真ん中がtabキー)0(ラベル)
すなわち、画像名tabキーラベル
注意ラベルは0から、ラベルは1からしないでください.そうしないと、エラーが発生します.
1.3最小データセット運転
直接dataフォルダに2つのファイルを入れます.jpg ,2.jpg
サブフォルダdata/Txtfileの2つのファイルtrain.txt, val.txt
どちらのファイルも

1.jpg	0
2.jpg	1

直接実行、ルートディレクトリ、python customData_train.py、正常に動作します.
1.4可能性のあるエラーおよび解決策(見ざるを得ないものはない)
pythonおよびtorchバージョンの問題によるエラー:

/home/xingxiangrui/env/lib/python3.6/site-packages/torchvision/transforms/transforms.py:563: UserWarning: The use of the transforms.RandomSizedCrop transform is deprecated, please use transforms.RandomResizedCrop instead.
"please use transforms.RandomResizedCrop instead.")
/home/xingxiangrui/env/lib/python3.6/site-packages/torchvision/transforms/transforms.py:188: UserWarning: The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.
"please use transforms.Resize instead.")

解決:2つの関数の名前を直接推奨される名前に置き換えます.

RuntimeError: cuda runtime error (59) : device-side assert triggered at /home/lychee/mycode/pytorch/aten/src/THC/generic/THCTensorMath.cu:24

解決:ラベルは0から、1からではなく、踊らないほうがいいです.

return loss.data[0]
IndexError: invalid index of a 0-dim tensor. Use tensor.item() to convert a 0-dim tensor to a Python 

解決:プログラム内の文を変更する
#原文:train_loss+=loss.data[0]#修正後:train_loss+=loss.item()
二、大サンプル量データセットの生成
2.1画像フォーマット

    print("Load dataset......")
    image_datasets = {x: customData(img_path='data/',
                                    txt_path=('data/TxtFile/' + x + '.txt'),
                                    data_transforms=data_transforms,
                                    dataset=x) for x in ['train', 'val']}

フォルダ名data/画像の保存
Data/TxtFileはtrainを保存する.txtとval.txt
2.2ラベル形式
train.txtにはトレーニングセットのピクチャ名とラベルが格納され、val.txtには検証セットの名前とラベルが格納されます.
ファイルのフォーマットは行ごとにデータ、ファイル名です.jpg(tabキー)ラベル番号

class customData(Dataset):
    def __init__(self, img_path, txt_path, dataset = '', data_transforms=None, loader = default_loader):
        with open(txt_path) as input_file:
            lines = input_file.readlines()
            self.img_name = [os.path.join(img_path, line.strip().split('\t')[0]) for line in lines]
            self.img_label = [int(line.strip().split('\t')[-1]) for line in lines]
        self.data_transforms = data_transforms
        self.dataset = dataset
        self.loader = loader

例:

1.jpg	0
2.jpg	1

2.3バッチ生成データとラベル
大量のデータはpython実操(二)を参照してデータセットを作成することができる:label変更とバッチ化ピクチャ処理
画像名読み出し

source_image_list = os.listdir(source_image_dir)
for idx in range(len(source_image_list)):
    if '.png' in source_image_list[idx-1]:
        continue
    elif '.jpg' in source_image_list[idx-1]:
        continue
    else:
        del source_image_list[idx]

注意、rangeは1から始まるので、後ろは-1にします.
シーヶンス乱れ
参照先:https://blog.csdn.net/amy_wumanfang/article/details/64483340
https://blog.csdn.net/matrix_google/article/details/72803741
直接 random.shuffle(list)， list で便利です.
例:

# -*- coding: utf-8 -*-
import random
#  list  ，  list    
list = range(10)
print list
random.shuffle(list)
print "       : ",  list

リストを2つのリストに分割
トレーニングセットと検証セットに分解し、それぞれ1/4と3/4

# train list and val list
source_train_list=[]
source_val_list=[]
for idx in range(len(source_image_list)):
    if idx

画像の読み出しと書き込み
画像はsrc_に保存されますimgでは、画像の名前を変更してsaveで書き込みます.

    # read dource images and rename
    path_source_img = os.path.join(source_image_dir, source_image_name)
    src_img = Image.open(path_source_img)
    full_image_name=prefix+"_train_"+source_image_name
    print(full_image_name)
    # save renamed image to the target dir
    target_image_path=os.path.join(target_image_dir, full_image_name)
    src_img.save(target_image_path)

ラベルの生成
作成ファイル:aは追加書き込みを示します

# create label_file or write label file
txt_file_train_name="train.txt"
txt_file_val_name="val.txt"
txt_file_train_path=os.path.join(txt_file_dir, txt_file_train_name)
txt_file_val_path=os.path.join(txt_file_dir, txt_file_val_name)
train_txt_file= open(txt_file_train_path,"a")
val_txt_file= open(txt_file_val_path,"a")

各行に「」を付ける必要がある

    # write image names and labels
    line_strings= full_image_name+"\t"+str(class_label)+"
"
    train_txt_file.write(line_strings)

三、訓練と検証
3.1データのロード
txtファイルの各行に直接基づいてデータとラベルをロードし、トレーニングを行うことができます.

    data_transforms = {
        'train': transforms.Compose([
            #transforms.RandomSizedCrop(224),
            transforms.RandomResizedCrop(224),
            transforms.RandomHorizontalFlip(),
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ]),
        'val': transforms.Compose([
            #transforms.Scale(256),
            transforms.Resize(256),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ]),
    }

    use_gpu = torch.cuda.is_available()

    batch_size = 32
    num_class = 3
    print("batch size:",batch_size,"num_classes:",num_class)

    print("Load dataset......")
    # image_datasets = {x: customData(img_path='sin_poly_defect_data/',
    #                                 txt_path=('sin_poly_defect_data/TxtFile/general_train.txt'),
    #                                 data_transforms=data_transforms,
    #                                 dataset=x) for x in ['train', 'total_val']}
    image_datasets={}
    image_datasets['train'] = customData(img_path='sin_poly_defect_data/',
                                         txt_path=('sin_poly_defect_data/TxtFile/general_train.txt'),
                                         data_transforms=data_transforms,
                                         dataset='train')
    image_datasets['val'] = customData(img_path='sin_poly_defect_data/',
                                       txt_path=('sin_poly_defect_data/TxtFile/real_poly_defect.txt'),
                                       data_transforms=data_transforms,
                                       dataset='val')
    # train_data=image_datasets.pop('general_train')
    # image_datasets['train']=train_data
    # val_data=image_datasets.pop('total_val')
    # image_datasets['val']=val_data

    # wrap your data and label into Tensor
    print("wrap data into Tensor......")
    dataloders = {x: torch.utils.data.DataLoader(image_datasets[x],
                                                 batch_size=batch_size,
                                                 shuffle=True) for x in ['train', 'val']}

    dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'val']}
    print("total dataset size:",dataset_sizes)

3.2データロード関数
データによる関数のロード

def default_loader(path):
    try:
        img = Image.open(path)
        return img.convert('RGB')
    except:
        print("Cannot read image: {}".format(path))

# define your Dataset. Assume each line in your .txt file is [name/tab/label], for example:0001.jpg 1
class customData(Dataset):
    def __init__(self, img_path, txt_path, dataset = '', data_transforms=None, loader = default_loader):
        with open(txt_path) as input_file:
            lines = input_file.readlines()
            self.img_name = [os.path.join(img_path, line.strip().split('\t')[0]) for line in lines]
            self.img_label = [int(line.strip().split('\t')[-1]) for line in lines]
        self.data_transforms = data_transforms
        self.dataset = dataset
        self.loader = loader

    def __len__(self):
        return len(self.img_name)

    def __getitem__(self, item):
        img_name = self.img_name[item]
        label = self.img_label[item]
        img = self.loader(img_name)

        if self.data_transforms is not None:
            try:
                img = self.data_transforms[self.dataset](img)
            except:
                print("Cannot transform image: {}".format(img_name))
        return img, label

3.3モデルトレーニングと検証
lossの定義

    print("Define loss function and optimizer......")
    # define cost function
    criterion = nn.CrossEntropyLoss()

    # Observe that all parameters are being optimized
    optimizer_ft = optim.SGD(model_ft.parameters(), lr=0.005, momentum=0.9)

    # Decay LR by a factor of 0.2 every 5 epochs
    exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=5, gamma=0.2)

    # multi-GPU
    model_ft = torch.nn.DataParallel(model_ft, device_ids=[0])

3.4モデルのトレーニングと保存

    # train model
    print("start train_model......")
    model_ft = train_model(model=model_ft,
                           criterion=criterion,
                           optimizer=optimizer_ft,
                           scheduler=exp_lr_scheduler,
                           num_epochs=15,
                           use_gpu=use_gpu)

    # save best model
    print("save model......")
    torch.save(model_ft,"output/resnet_on_PV_best_total_val.pkl")

四、コード
githubアドレス:https://github.com/Xingxiangrui/image_classification_with_pytorch

# -*- coding: utf-8 -*
"""
Created by Xingxiangrui on 2019.5.9
This code is to :
    1. copy image from source_image_dir to the target_image_dir
    2. And generate .txt file for further training
        in which each line is : image_name.jpg  (tab)  image_label (from 0)
        such as:
            image_01.jpg    0
            iamge_02.jpg    1
            ...
            image_02.jpg    0

"""

# import matplotlib.pyplot as plt
import numpy as np
from PIL import Image
import os
import random

# variables need to be change
source_image_dir="/Users/Desktop/used/SuZhouRuiTu_dataset/single-poly-defect/poly_OK"
target_image_dir="/Users/Desktop/used/SuZhouRuiTu_dataset/data_for_resnet_classification"
txt_file_dir="/Users/Desktop/used/SuZhouRuiTu_dataset/data_for_resnet_classification/TxtFile"
prefix="poly_OK"
class_label=1
# label 0: single_OK ; label_1: poly_OK ; label 2: poly_defect

print("Program Start......")
print("-"*20)
print("-"*20)
print("-"*20)

# load image list in the source dir
source_image_list = os.listdir(source_image_dir)
for idx in range(len(source_image_list)):
    if '.png' in source_image_list[idx-1]:
        continue
    elif '.jpg' in source_image_list[idx-1]:
        continue
    else:
        del source_image_list[idx-1]

# shuffle image list
print("initial list:")
print source_image_list
random.shuffle(source_image_list)
print("shuffled list:")
print source_image_list

# train list and val list
source_train_list=[]
source_val_list=[]
for idx in range(len(source_image_list)):
    if idx

# -*- coding: utf-8 -*
"""
created by xingxiangrui on 2019.5.9
This is the pytorch code for iamge classification
python 3.6 and torch 0.4.1 is ok

dataset mode:
    folder data in which is jpg images
    folder data/TxtFile/ in which is train.txt and val.txt
        in train.txt each line :
        image_name.jpg  (tab)  image_label (from 0)
        such as:
        image_01.jpg    0
        iamge_02.jpg    1
        ...
        image_02.jpg    0

"""
from __future__ import print_function, division
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
from torch.autograd import Variable
from torchvision import  models, transforms
import time
import os
from torch.utils.data import Dataset
from PIL import Image

# use PIL Image to read image
def default_loader(path):
    try:
        img = Image.open(path)
        return img.convert('RGB')
    except:
        print("Cannot read image: {}".format(path))

# define your Dataset. Assume each line in your .txt file is [name/tab/label], for example:0001.jpg 1
class customData(Dataset):
    def __init__(self, img_path, txt_path, dataset = '', data_transforms=None, loader = default_loader):
        with open(txt_path) as input_file:
            lines = input_file.readlines()
            self.img_name = [os.path.join(img_path, line.strip().split('\t')[0]) for line in lines]
            self.img_label = [int(line.strip().split('\t')[-1]) for line in lines]
        self.data_transforms = data_transforms
        self.dataset = dataset
        self.loader = loader

    def __len__(self):
        return len(self.img_name)

    def __getitem__(self, item):
        img_name = self.img_name[item]
        label = self.img_label[item]
        img = self.loader(img_name)

        if self.data_transforms is not None:
            try:
                img = self.data_transforms[self.dataset](img)
            except:
                print("Cannot transform image: {}".format(img_name))
        return img, label

def train_model(model, criterion, optimizer, scheduler, num_epochs, use_gpu):
    since = time.time()

    best_model_wts = model.state_dict()
    best_acc = 0.0

    for epoch in range(num_epochs):
        begin_time = time.time()
        print('Epoch {}/{}'.format(epoch, num_epochs - 1))
        print('-' * 10)

        # Each epoch has a training and validation phase
        for phase in ['train', 'val']:
            count_batch = 0
            if phase == 'train':
                scheduler.step()
                model.train(True)  # Set model to training mode
            else:
                model.train(False)  # Set model to evaluate mode

            running_loss = 0.0
            running_corrects = 0.0

            # Iterate over data.
            for data in dataloders[phase]:
                count_batch += 1
                # get the inputs
                inputs, labels = data

                # wrap them in Variable
                if use_gpu:
                    inputs = Variable(inputs.cuda())
                    labels = Variable(labels.cuda())
                else:
                    inputs, labels = Variable(inputs), Variable(labels)

                # zero the parameter gradients
                optimizer.zero_grad()

                # forward
                outputs = model(inputs)
                _, preds = torch.max(outputs.data, 1)
                loss = criterion(outputs, labels)

                # backward + optimize only if in training phase
                if phase == 'train':
                    loss.backward()
                    optimizer.step()
                # statistics
                running_loss += loss.item()
                running_corrects += torch.sum(preds == labels.data).to(torch.float32)

                # print result every 10 batch
                if count_batch%10 == 0:
                    batch_loss = running_loss / (batch_size*count_batch)
                    batch_acc = running_corrects / (batch_size*count_batch)
                    print('{} Epoch [{}] Batch [{}] Loss: {:.4f} Acc: {:.4f} Time: {:.4f}s'. \
                          format(phase, epoch, count_batch, batch_loss, batch_acc, time.time()-begin_time))
                    begin_time = time.time()

            epoch_loss = running_loss / dataset_sizes[phase]
            epoch_acc = running_corrects / dataset_sizes[phase]
            print('{} Loss: {:.4f} Acc: {:.4f}'.format(phase, epoch_loss, epoch_acc))

            # save model
            if phase == 'train':
                if not os.path.exists('output'):
                    os.makedirs('output')
                torch.save(model, 'output/resnet_on_PV_epoch{}.pkl'.format(epoch))

            # deep copy the model
            if phase == 'val' and epoch_acc > best_acc:
                best_acc = epoch_acc
                best_model_wts = model.state_dict()

    time_elapsed = time.time() - since
    print('Training complete in {:.0f}m {:.0f}s'.format(
        time_elapsed // 60, time_elapsed % 60))
    print('Best val Acc: {:4f}'.format(best_acc))

    # load best model weights
    model.load_state_dict(best_model_wts)
    return model

if __name__ == '__main__':
    print("Program start","-"*10)

    print("Init data transforms......")
    data_transforms = {
        'train': transforms.Compose([
            #transforms.RandomSizedCrop(224),
            transforms.RandomResizedCrop(224),
            transforms.RandomHorizontalFlip(),
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ]),
        'val': transforms.Compose([
            #transforms.Scale(256),
            transforms.Resize(256),
            transforms.CenterCrop(224),
            transforms.ToTensor(),
            transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
        ]),
    }

    use_gpu = torch.cuda.is_available()

    batch_size = 32
    num_class = 3
    print("batch size:",batch_size,"num_classes:",num_class)

    print("Load dataset......")
    # image_datasets = {x: customData(img_path='sin_poly_defect_data/',
    #                                 txt_path=('sin_poly_defect_data/TxtFile/general_train.txt'),
    #                                 data_transforms=data_transforms,
    #                                 dataset=x) for x in ['train', 'total_val']}
    image_datasets={}
    image_datasets['train'] = customData(img_path='sin_poly_defect_data/',
                                         txt_path=('sin_poly_defect_data/TxtFile/general_train.txt'),
                                         data_transforms=data_transforms,
                                         dataset='train')
    image_datasets['val'] = customData(img_path='sin_poly_defect_data/',
                                       txt_path=('sin_poly_defect_data/TxtFile/real_poly_defect.txt'),
                                       data_transforms=data_transforms,
                                       dataset='val')
    # train_data=image_datasets.pop('general_train')
    # image_datasets['train']=train_data
    # val_data=image_datasets.pop('total_val')
    # image_datasets['val']=val_data

    # wrap your data and label into Tensor
    print("wrap data into Tensor......")
    dataloders = {x: torch.utils.data.DataLoader(image_datasets[x],
                                                 batch_size=batch_size,
                                                 shuffle=True) for x in ['train', 'val']}

    dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'val']}
    print("total dataset size:",dataset_sizes)

    # get model and replace the original fc layer with your fc layer
    print("get resnet model and replace last fc layer...")
    model_ft = models.resnet50(pretrained=True)
    num_ftrs = model_ft.fc.in_features
    model_ft.fc = nn.Linear(num_ftrs, num_class)

    # if use gpu
    print("Use gpu:",use_gpu)
    if use_gpu:
        model_ft = model_ft.cuda()

    print("Define loss function and optimizer......")
    # define cost function
    criterion = nn.CrossEntropyLoss()

    # Observe that all parameters are being optimized
    optimizer_ft = optim.SGD(model_ft.parameters(), lr=0.005, momentum=0.9)

    # Decay LR by a factor of 0.2 every 5 epochs
    exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=5, gamma=0.2)

    # multi-GPU
    model_ft = torch.nn.DataParallel(model_ft, device_ids=[0])

    # train model
    print("start train_model......")
    model_ft = train_model(model=model_ft,
                           criterion=criterion,
                           optimizer=optimizer_ft,
                           scheduler=exp_lr_scheduler,
                           num_epochs=15,
                           use_gpu=use_gpu)
    # save best model
    print("save model......")
    torch.save(model_ft,"output/resnet_on_PV_best_total_val.pkl")