目录
猫狗大战--使用 “VGG16进行CIFAR10分类” 迁移学习实现
一、在colab上使用数据集
二、训练模型
三、测试数据 Valid(研习社的test在下一部分)
四、研习社测试Test
以下为旧版本 2020年11月17日 13点48分
使用VGG模型进行猫狗大战
一、代码部分
猫狗大战训练代码.ipynb
https://colab.research.google.com/drive/1qbo216iUiKvfnwNtVcZ2fFH821VzJoIu?usp=sharing
猫狗大战测试代码.ipynb
https://colab.research.google.com/drive/1Ulwkgn87dzDeu4nrfdIEEsyZ1YCth9Eo?usp=sharing
最优模型(Model) 百度云盘地址
链接:https://pan.baidu.com/s/1jbcO3UYiPDYRSrHpM7B7Qg 提取码:1yk2
有两种方案:
! wget *url*
from google.colab import drive
drive.mount('/content/drive')
#具体看你自己的文件放在哪
#1、如果是直接下载的 就默认在 conten/ 下面
#2、用drive里的文件就在下面的路径中,可以点击左边的文件目录查看
!unzip "/content/drive/My Drive/data.zip"
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
vgg_format = transforms.Compose([
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])
data_dir = '/content/data'
dsets = {x: datasets.ImageFolder(os.path.join(data_dir, x), vgg_format)
for x in ['train']}
dset_sizes = {x: len(dsets[x]) for x in ['train']}
dset_classes = dsets['train'].classes
#另一个数据集
# data_dir2 = '/content/dogscats'
# dsets2 = {x: datasets.ImageFolder(os.path.join(data_dir2, x), vgg_format)
# for x in ['valid']}
# dset_sizes2 = {x: len(dsets2[x]) for x in ['valid']}
# dset_classes2 = dsets2['valid'].classes
# 通过下面代码可以查看 dsets 的一些属性
print(dsets['train'].classes)
print(dsets['train'].class_to_idx)
print(dsets['train'].imgs[:5])
print('dset_sizes: ', dset_sizes)
loader_train = torch.utils.data.DataLoader(dsets['train'], batch_size=256, shuffle=True, num_workers=6)
# loader_valid = torch.utils.data.DataLoader(dsets2['valid'], batch_size=5, shuffle=False, num_workers=6)
# count = 1
# for data in loader_valid:
# print(count, end='\n')
# if count == 1:
# inputs_try,labels_try = data
# count +=1
# print(labels_try)
# print(inputs_try.shape)
注释掉的是验证集中的数据,因为这部分主要是训练,测试和验证我会跟训练分开,原因下面会介绍。
#下载
!wget https://s3.amazonaws.com/deep-learning-models/image-models/imagenet_class_index.json
如果下面的运行不出来,把上面的验证集注释取消就可以了,也可以把下面报错的行全 删掉/注释 掉。
#使用vgg16需要
model_vgg = models.vgg16(pretrained=True)
with open('./imagenet_class_index.json') as f:
class_dict = json.load(f)
dic_imagenet = [class_dict[str(i)][1] for i in range(len(class_dict))]
inputs_try , labels_try = inputs_try.to(device), labels_try.to(device)
model_vgg = model_vgg.to(device)
outputs_try = model_vgg(inputs_try)
print(outputs_try)
print(outputs_try.shape)
'''
可以看到结果为5行,1000列的数据,每一列代表对每一种目标识别的结果。
但是我也可以观察到,结果非常奇葩,有负数,有正数,
为了将VGG网络输出的结果转化为对每一类的预测概率,我们把结果输入到 Softmax 函数
'''
m_softm = nn.Softmax(dim=1)
probs = m_softm(outputs_try)
vals_try,pred_try = torch.max(probs,dim=1)
print( 'prob sum: ', torch.sum(probs,1))
print( 'vals_try: ', vals_try)
print( 'pred_try: ', pred_try)
def imshow(inp, title=None):
# Imshow for Tensor.
inp = inp.numpy().transpose((1, 2, 0))
mean = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])
inp = np.clip(std * inp + mean, 0,1)
plt.imshow(inp)
if title is not None:
plt.title(title)
plt.pause(0.001) # pause a bit so that plots are updated
print([dic_imagenet[i] for i in pred_try.data])
imshow(torchvision.utils.make_grid(inputs_try.data.cpu()),
title=[dset_classes[x] for x in labels_try.data.cpu()])
print(model_vgg)
model_vgg_new = model_vgg;
#冻结VGG16中的参数,不进行梯度下降
for param in model_vgg_new.parameters():
param.requires_grad = False
#新增两个线性层,后期主要训练这两层
model_vgg_new.classifier._modules['6'] = nn.Linear(4096, 4096)
model_vgg_new.classifier._modules['7'] = nn.ReLU(inplace=False)
model_vgg_new.classifier._modules['8'] = nn.Dropout(p=0.5,inplace=False)
model_vgg_new.classifier._modules['9'] = nn.Linear(4096, 2)
model_vgg_new.classifier._modules['10'] = torch.nn.LogSoftmax(dim=1)
model_vgg_new = model_vgg_new.to(device)
print(model_vgg_new.classifier)
训练模型
① 我把SGD改成了Adam;from tqdm import trange,tqdm
criterion = nn.NLLLoss()
lr = 0.001
optimizer_vgg = torch.optim.Adam(model_vgg_new.classifier[6].parameters(), lr=lr)
def train_model(model, dataloader, size, epochs=200, optimizer=None):
model.train()
max_acc = 0
count = 0
for epoch in range(epochs):
running_loss = 0.0
running_corrects = 0
count = 0
for inputs, classes in tqdm(dataloader):
inputs = inputs.to(device)
classes = classes.to(device)
outputs = model(inputs)
loss = criterion(outputs, classes)
optimizer = optimizer
optimizer.zero_grad()
loss.backward()
optimizer.step()
_, preds = torch.max(outputs.data, 1)
# statistics
running_loss += loss.data.item()
running_corrects += torch.sum(preds == classes.data)
count += len(inputs)
#print('Training: No. ', count, ' process ... total: ', size)
epoch_loss = running_loss / size
epoch_acc = running_corrects.data.item() / size
if epoch_acc>max_acc:
max_acc = epoch_acc
torch.save(model, '/content/drive/My Drive/model_best_new.pth')
tqdm.write("\n Got A Nice Model Acc:{:.8f}".format(max_acc))
tqdm.write('\nepoch: {} \tLoss: {:.8f} Acc: {:.8f}'.format(epoch,epoch_loss, epoch_acc))
time.sleep(0.1)
torch.save(model, '/content/drive/My Drive/model_last_new.pth')
tqdm.write("Got A Nice Model")
# 模型训练
train_model(model_vgg_new, loader_train, size=dset_sizes["train"], epochs=100,
optimizer=optimizer_vgg)
在上一部分,我将train和valid分开是原因的是:
from google.colab import drive
drive.mount('/content/drive')
#视具体情况而定
!unzip "/content/drive/My Drive/test.zip"
import os
import torch
from torchvision import transforms,datasets
from tqdm import tqdm
device = torch.device("cuda:0")
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
vgg_format = transforms.Compose([
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])
#注意这里的文件夹名称,我的是test,因为我的压缩包就叫test
data_dir = r'test'
file_name = 'valid'#"train1"
dsets = {x: datasets.ImageFolder(os.path.join(data_dir, x), vgg_format)
for x in [file_name]}
dset_sizes = {x: len(dsets[x]) for x in [file_name]}
loader_valid = torch.utils.data.DataLoader(dsets[file_name], batch_size=256, shuffle=False, num_workers=0)
#这里是需要记载的模型
model_vgg_new = torch.load(r'/content/drive/My Drive/model_best_new.pth')
model_vgg_new = model_vgg_new.to(device)
def test_model(model,dataloader,size):
model.eval()
running_corrects = 0
for inputs,classes in tqdm(dataloader):
inputs = inputs.to(device)
classes = classes.to(device)
outputs = model(inputs)
_,preds = torch.max(outputs.data,1)
running_corrects += torch.sum(preds == classes.data)
epoch_acc = running_corrects.data.item() / size
tqdm.write('Acc: {:.4f} '.format(epoch_acc))
test_model(model_vgg_new, loader_valid, size=dset_sizes[file_name])
import torch
import numpy as np
from torchvision import transforms,datasets
from tqdm import tqdm
device = torch.device("cuda:0" )
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
vgg_format = transforms.Compose([
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])
#注意这里,我的数据在yanxishe这个文件夹里
dsets_mine = datasets.ImageFolder(r"yanxishe", vgg_format)
loader_test = torch.utils.data.DataLoader(dsets_mine, batch_size=1, shuffle=False, num_workers=0)
#模型的具体地址需要根据具体情况修改
model_vgg_new = torch.load(r'/content/drive/My Drive/model_best_16.pth')
model_vgg_new = model_vgg_new.to(device)
dic = {}
def test(model,dataloader,size):
model.eval()
predictions = np.zeros(size)
cnt = 0
for inputs,_ in tqdm(dataloader):
inputs = inputs.to(device)
outputs = model(inputs)
_,preds = torch.max(outputs.data,1)
#这里是切割路径,因为dset中的数据不是按1-2000顺序排列的
key = dsets_mine.imgs[cnt][0].split("\\")[-1].split('.')[0]
dic[key] = preds[0]
cnt = cnt +1
test(model_vgg_new,loader_test,size=2000)
with open("result18.csv",'a+') as f:
for key in range(2000):
#这里的yanxishe/test/是我的图片路径,按需更换
f.write("{},{}\n".format(key,dic["yanxishe/test/"+str(key)]))
使用VGG模型进行猫狗大战
原文见:https://github.com/mlelarge/dataflowr/blob/master/CEA_EDF_INRIA/01_intro_DLDIY_colab.ipynb
VGG是由Simonyan 和Zisserman在文献《Very Deep Convolutional Networks for Large Scale Image Recognition》中提出卷积神经网络模型,其名称来源于作者所在的牛津大学视觉几何组(Visual Geometry Group)的缩写。该模型参加2014年的 ImageNet图像分类与定位挑战赛,取得了优异成绩:在分类任务上排名第二,在定位任务上排名第一。
迁移学习是一种机器学习方法,就是把为任务 A 开发的模型作为初始点,重新使用在为任务 B 开发模型的过程中。
ImageNet图像分类10类中存在猫和狗,所以用VGG来作为“猫狗大战”的预训练是十分合理的
import os
import torch
import torch.nn as nn
from torchvision import models,transforms,datasets
from tqdm import trange,tqdm
# 判断是否存在GPU设备
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print('Using gpu: %s ' % torch.cuda.is_available())
datasets 是 torchvision 中的一个包,可以用做加载图像数据。它可以以多线程(multi-thread)的形式从硬盘中读取数据,使用 mini-batch 的形式,在网络训练中向 GPU 输送。在使用CNN处理图像时,需要进行预处理。图片将被整理成 224*224*3 的大小,同时还将进行归一化处理。
这里我将https://static.leiphone.com/cat_dog.rar的训练文件一起加入了到了训练集中,因为原来的训练集只有1800张图片,我希望能在更大的数据集上进行训练,以求获得更好的结果。
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
vgg_format = transforms.Compose([
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])
data_dir = r'G:\作业\#1.人工智能\colab_demo-master\dogscats'
dsets = {x: datasets.ImageFolder(os.path.join(data_dir, x), vgg_format)
for x in ['train', 'valid']}
dset_sizes = {x: len(dsets[x]) for x in ['train', 'valid']}
dset_classes = dsets['train'].classes
loader_train = torch.utils.data.DataLoader(dsets['train'], batch_size=40, shuffle=True, num_workers=0)
loader_valid = torch.utils.data.DataLoader(dsets['valid'], batch_size=16, shuffle=False, num_workers=0)
这里我把训练出来比较好的模型序列化到硬盘上
首先,准确率比较高的情况往往不是训练的最终结果,选择准确率较高的模型保存到本地可以获得比较好的效果
其次,2w+数据的训练周期较长,拿出其中表现较好时刻的模型可以提前进行测试,提高效率
n = input("是否重新训练?(Y/N)")
if n=='N':
path = input("输入文件名:")
CNT = input("输入轮数:")
model_vgg_new = torch.load(path)
model_vgg_new = model_vgg_new.to(device)
else:
model_vgg = models.vgg16(pretrained=True)
model_vgg = model_vgg.to(device)
model_vgg_new = model_vgg
for param in model_vgg_new.parameters():
param.requires_grad = False
model_vgg_new.classifier._modules['6'] = nn.Linear(4096, 2)
model_vgg_new.classifier._modules['7'] = torch.nn.LogSoftmax(dim = 1)
将表现较好时刻的模型存盘到本地
model_vgg_new = model_vgg_new.to(device)
criterion = nn.NLLLoss()
# 学习率
lr = 0.001
optimizer_vgg = torch.optim.Adam(model_vgg_new.classifier[6].parameters(), lr=lr)
'''
第二步:训练模型
'''
N_ = int(CNT)+1
def train_model(model, dataloader, size, epochs=100, optimizer=None):
model.train()
global N_
for epoch in range(epochs):
running_loss = 0.0
running_corrects = 0
count = 0
for inputs, classes in tqdm(dataloader):
inputs = inputs.to(device)
classes = classes.to(device)
outputs = model(inputs)
loss = criterion(outputs, classes)
optimizer = optimizer
optimizer.zero_grad()
loss.backward()
optimizer.step()
_, preds = torch.max(outputs.data, 1)
# statistics
running_loss += loss.data.item()
running_corrects += torch.sum(preds == classes.data)
count += len(inputs)
#print('Training: No. ', count, ' process ... total: ', size)
epoch_loss = running_loss / size
epoch_acc = running_corrects.data.item() / size
print('{} \tLoss: {:.4f} Acc: {:.4f}'.format(N_,epoch_loss, epoch_acc))
if epoch_acc > 0.97:
torch.save(model, './model'+str(N_)+'_'+str(epoch_acc)+'_'+'.pth')
print("Got A Nice Model")
N_ = N_ + 1
# 模型训练
train_model(model_vgg_new, loader_train, size=dset_sizes['train'], epochs=100,
optimizer=optimizer_vgg)
import os
import torch
from torchvision import transforms,datasets
from tqdm import tqdm
device = torch.device("cuda:0")
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
vgg_format = transforms.Compose([
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])
data_dir = r'G:\作业\#1.人工智能\colab_demo-master\dogscats'
dsets = {x: datasets.ImageFolder(os.path.join(data_dir, x), vgg_format)
for x in ['valid']}
dset_sizes = {x: len(dsets[x]) for x in ['valid']}
loader_valid = torch.utils.data.DataLoader(dsets['valid'], batch_size=8, shuffle=False, num_workers=0)
model_vgg_new = torch.load('xxxxxx')
model_vgg_new = model_vgg_new.to(device)
def test_model(model,dataloader,size):
model.eval()
running_corrects = 0
for inputs,classes in tqdm(dataloader):
inputs = inputs.to(device)
classes = classes.to(device)
outputs = model(inputs)
_,preds = torch.max(outputs.data,1)
running_corrects += torch.sum(preds == classes.data)
epoch_acc = running_corrects.data.item() / size
print('Acc: {:.4f} '.format(epoch_acc))
test_model(model_vgg_new, loader_valid, size=dset_sizes['valid'])
import torch
import numpy as np
from torchvision import transforms,datasets
from tqdm import tqdm
device = torch.device("cuda:0" )
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
vgg_format = transforms.Compose([
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])
dsets_mine = datasets.ImageFolder(r"G:\作业\#1.人工智能\colab_demo-master\dogscats2", vgg_format)
loader_test = torch.utils.data.DataLoader(dsets_mine, batch_size=1, shuffle=False, num_workers=0)
model_vgg_new = torch.load('')
model_vgg_new = model_vgg_new.to(device)
dic = {}
def test(model,dataloader,size):
model.eval()
predictions = np.zeros(size)
cnt = 0
for inputs,_ in tqdm(dataloader):
inputs = inputs.to(device)
outputs = model(inputs)
_,preds = torch.max(outputs.data,1)
key = dsets_mine.imgs[cnt][0].split("\\")[-1].split('.')[0]
dic[key] = preds[0]
cnt = cnt +1
test(model_vgg_new,loader_test,size=2000)
with open("result.csv",'a+') as f:
for key in range(2000):
f.write("{},{}\n".format(key,dic[str(key)]))