使用pytorch实现基于VGG 19预训练模型的鲜花识别分类器

项目说明

本文使用的数据集是网络开源的鲜花数据集，并且基于VGG19的预训练模型通过迁移学习重新训练鲜花数据由此构建一个鲜花识别分类器

数据集

可以在此处找到有关花朵数据集的信息。数据集为102个花类的每一个都包含一个单独的文件夹。每朵花都标记为一个数字，每个编号的目录都包含许多.jpg文件。

实验环境

prtorch库

PIL库

如果想使用GPU训练的话请使用英伟达的显卡并安装好CUDA

如果用GPU的话我在自己电脑上使用GPU只使用了91分钟（我的GPU是1050）

倒入库并检测是否有可用GPU

%matplotlib inline

%config InlineBackend.figure_format = 'retina'

import time

import json

import copy

import matplotlib.pyplot as plt

import seaborn as sns

import numpy as np

import PIL

from PIL import Image

from collections import OrderedDict

import torch

from torch import nn, optim

from torch.optim import lr_scheduler

from torch.autograd import Variable

import torchvision

from torchvision import datasets, models, transforms

from torch.utils.data.sampler import SubsetRandomSampler

import torch.nn as nn

import torch.nn.functional as F

import os

# check if GPU is available

train_on_gpu = torch.cuda.is_available()

if not train_on_gpu:

    print('Bummer!  Training on CPU ...')

else:

    print('You are good to go!  Training on GPU ...')

有GPU就启用

    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

定义数据集位置

data_dir = 'F:\资料\项目\image_classifier_pytorch-master\\flower_data'

train_dir = 'flower_data/train'

valid_dir = 'flower_data/valid'

导入数据集并对数据进行处理

# Define your transforms for the training and testing sets

data_transforms = {

    'train': transforms.Compose([

        transforms.RandomRotation(30),

        transforms.RandomResizedCrop(224),

        transforms.RandomHorizontalFlip(),

        transforms.ToTensor(),

        transforms.Normalize([0.485, 0.456, 0.406],

                             [0.229, 0.224, 0.225])

    ]),

    'valid': transforms.Compose([

        transforms.Resize(256),

        transforms.CenterCrop(224),

        transforms.ToTensor(),

        transforms.Normalize([0.485, 0.456, 0.406],

                             [0.229, 0.224, 0.225])

    ])

}

# Load the datasets with ImageFolder

image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x),

                                          data_transforms[x])

                  for x in ['train', 'valid']}

# Using the image datasets and the trainforms, define the dataloaders

batch_size = 64

dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=batch_size,

                                             shuffle=True, num_workers=4)

              for x in ['train', 'valid']}

class_names = image_datasets['train'].classes

dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'valid']}

class_names = image_datasets['train'].classes

# Label mapping

with open('F:\资料\项目\image_classifier_pytorch-master\cat_to_name.json', 'r') as f:

    cat_to_name = json.load(f)

查看数据情况

# Run this to test the data loader

images, labels = next(iter(dataloaders['train']))

images.size()

# # Run this to test your data loader

images, labels = next(iter(dataloaders['train']))

rand_idx = np.random.randint(len(images))

# print(rand_idx)

print("label: {}, class: {}, name: {}".format(labels[rand_idx].item(),

                                               class_names[labels[rand_idx].item()],

                                               cat_to_name[class_names[labels[rand_idx].item()]]))

定义模型

model_name = 'densenet' #vgg

if model_name == 'densenet':

    model = models.densenet161(pretrained=True)

    num_in_features = 2208

    print(model)

elif model_name == 'vgg':

    model = models.vgg19(pretrained=True)

    num_in_features = 25088

    print(model.classifier)

else:

    print("Unknown model, please choose 'densenet' or 'vgg'")

# Create classifier

for param in model.parameters():

    param.requires_grad = False

def build_classifier(num_in_features, hidden_layers, num_out_features):

    classifier = nn.Sequential()

    if hidden_layers == None:

        classifier.add_module('fc0', nn.Linear(num_in_features, 102))

    else:

        layer_sizes = zip(hidden_layers[:-1], hidden_layers[1:])

        classifier.add_module('fc0', nn.Linear(num_in_features, hidden_layers[0]))

        classifier.add_module('relu0', nn.ReLU())

function(){ //XM返佣 http://www.kaifx.cn/broker/xm.html

        classifier.add_module('drop0', nn.Dropout(.6))

        classifier.add_module('relu1', nn.ReLU())

        classifier.add_module('drop1', nn.Dropout(.5))

        for i, (h1, h2) in enumerate(layer_sizes):

            classifier.add_module('fc'+str(i+1), nn.Linear(h1, h2))

            classifier.add_module('relu'+str(i+1), nn.ReLU())

            classifier.add_module('drop'+str(i+1), nn.Dropout(.5))

        classifier.add_module('output', nn.Linear(hidden_layers[-1], num_out_features))

    return classifier

hidden_layers = None#[4096, 1024, 256][512, 256, 128]

classifier = build_classifier(num_in_features, hidden_layers, 102)

print(classifier)

 # Only train the classifier parameters, feature parameters are frozen

if model_name == 'densenet':

    model.classifier = classifier

    criterion = nn.CrossEntropyLoss()

    optimizer = optim.Adadelta(model.parameters()) # Adadelta #weight optim.Adam(model.parameters(), lr=0.001, momentum=0.9)

    #optimizer_conv = optim.SGD(model.parameters(), lr=0.0001, weight_decay=0.001, momentum=0.9)

    sched = optim.lr_scheduler.StepLR(optimizer, step_size=4)

elif model_name == 'vgg':

    model.classifier = classifier

    criterion = nn.NLLLoss()

    optimizer = optim.Adam(model.classifier.parameters(), lr=0.0001)

    sched = lr_scheduler.StepLR(optimizer, step_size=4, gamma=0.1)

else:

    pass

def train_model(model, criterion, optimizer, sched, num_epochs=5):

    since = time.time()

    best_model_wts = copy.deepcopy(model.state_dict())

    best_acc = 0.0

    for epoch in range(num_epochs):

        print('Epoch {}/{}'.format(epoch+1, num_epochs))

        print('-' * 10)

        # Each epoch has a training and validation phase

        for phase in ['train', 'valid']:

            if phase == 'train':

                model.train()  # Set model to training mode

            else:

                model.eval()   # Set model to evaluate mode

            running_loss = 0.0

            running_corrects = 0

            # Iterate over data.

            for inputs, labels in dataloaders[phase]:

                inputs = inputs.to(device)

                labels = labels.to(device)

                # zero the parameter gradients

                optimizer.zero_grad()

                # forward

                # track history if only in train

                with torch.set_grad_enabled(phase == 'train'):

                    outputs = model(inputs)

                    _, preds = torch.max(outputs, 1)

                    loss = criterion(outputs, labels)

                    # backward + optimize only if in training phase

                    if phase == 'train':

                        #sched.step()

                        loss.backward()

                        optimizer.step()

                # statistics

                running_loss += loss.item() * inputs.size(0)

                running_corrects += torch.sum(preds == labels.data)

            epoch_loss = running_loss / dataset_sizes[phase]

            epoch_acc = running_corrects.double() / dataset_sizes[phase]

            print('{} Loss: {:.4f} Acc: {:.4f}'.format(

                phase, epoch_loss, epoch_acc))

            # deep copy the model

            if phase == 'valid' and epoch_acc > best_acc:

                best_acc = epoch_acc

                best_model_wts = copy.deepcopy(model.state_dict())

        print()

    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

开始训练

epochs = 30

model.to(device)

model = train_model(model, criterion, optimizer, sched, epochs)