直接搭建網(wǎng)絡(luò)必須與torchvision自帶的網(wǎng)絡(luò)的權(quán)重也就是pth文件的結(jié)構(gòu)、尺寸和變量命名完全一致，否則無(wú)法加載權(quán)重文件。
此時(shí)可比較2個(gè)字典逐一加載，詳見
pytorch加載預(yù)訓(xùn)練模型與自己模型不匹配的解決方案
import torch
import torchvision
import cv2 as cv
from utils.utils import letter_box
from model.backbone import ResNet18
model1 = ResNet18(1)
model2 = torchvision.models.resnet18(progress=False)
fc = model2.fc
model2.fc = torch.nn.Linear(512, 1)
# print(model)
model_dict1 = model1.state_dict()
model_dict2 = torch.load('resnet18.pth')
model_list1 = list(model_dict1.keys())
model_list2 = list(model_dict2.keys())
len1 = len(model_list1)
len2 = len(model_list2)
minlen = min(len1, len2)
for n in range(minlen):
if model_dict1[model_list1[n]].shape != model_dict2[model_list2[n]].shape:
continue
model_dict1[model_list1[n]] = model_dict2[model_list2[n]]
model1.load_state_dict(model_dict1)
missing, unspected = model2.load_state_dict(model_dict2)
image = cv.imread('zhn1.jpg')
image = letter_box(image, 224)
image = image[:, :, ::-1].transpose(2, 0, 1)
print('Network loading complete.')
model1.eval()
model2.eval()
with torch.no_grad():
image = torch.tensor(image/256, dtype=torch.float32).unsqueeze(0)
predict1 = model1(image)
predict2 = model2(image)
print('finished')
# torch.save(model.state_dict(), 'resnet18.pth')
以上為全部程序，最終可測(cè)試原模型與加載了自帶權(quán)重的自定義模型的輸出是否相等。
補(bǔ)充：使用Pytorch搭建ResNet分類網(wǎng)絡(luò)并基于遷移學(xué)習(xí)訓(xùn)練
如果stride=1，padding=1
卷積處理是不會(huì)改變特征矩陣的高和寬
使用BN層時(shí)
卷積中的參數(shù)bias置為False(有無(wú)偏置BN層的輸出都相同)，BN層放在conv層和relu層的中間
復(fù)習(xí)BN層：
Batch Norm 層是對(duì)每層數(shù)據(jù)歸一化后再進(jìn)行線性變換改善數(shù)據(jù)分布, 其中的線性變換是可學(xué)習(xí)的.
Batch Norm優(yōu)點(diǎn):減輕過(guò)擬合;改善梯度傳播（權(quán)重不會(huì)過(guò)高或過(guò)低）容許較高的學(xué)習(xí)率，能夠提高訓(xùn)練速度。減輕對(duì)初始化權(quán)重的強(qiáng)依賴，使得數(shù)據(jù)分布在激活函數(shù)的非飽和區(qū)域，一定程度上解決梯度消失問(wèn)題。作為一種正則化的方式，在某種程度上減少對(duì)dropout的使用。
Batch Norm層擺放位置:在激活層（如 ReLU ）之前還是之后，沒有一個(gè)統(tǒng)一的定論。
BN層與 Dropout 合作：Batch Norm的提出使得dropout的使用減少,但是Batch Norm不能完全取代dropout,保留較小的dropout率,如0.2可能效果更佳。
為什么要先normalize再通過(guò)γ,β線性變換恢復(fù)接近原來(lái)的樣子，這不是多此一舉嗎？
在一定條件下可以糾正原始數(shù)據(jù)的分布（方差，均值變?yōu)樾轮郸?β），當(dāng)原始數(shù)據(jù)分布足夠好時(shí)就是恒等映射，不改變分布。如果不做BN，方差和均值對(duì)前面網(wǎng)絡(luò)的參數(shù)有復(fù)雜的關(guān)聯(lián)依賴，具有復(fù)雜的非線性。在新參數(shù) γH′ + β 中僅由 γ,β 確定，與前邊網(wǎng)絡(luò)的參數(shù)無(wú)關(guān)，因此新參數(shù)很容易通過(guò)梯度下降來(lái)學(xué)習(xí)，能夠?qū)W習(xí)到較好的分布。
遷移學(xué)習(xí)導(dǎo)入權(quán)重和下載權(quán)重：
import torchvision.models.resnet#ctrl+鼠標(biāo)左鍵點(diǎn)擊即可下載權(quán)重
net = resnet34()#一開始不能設(shè)置全連接層的輸出種類為自己想要的，必須先將模型參數(shù)載入，再修改全連接層
# 官方提供載入預(yù)訓(xùn)練模型的方法
model_weight_path = "./resnet34-pre.pth"#權(quán)重路徑
missing_keys, unexpected_keys = net.load_state_dict(torch.load(model_weight_path), strict=False)#載入模型權(quán)重
inchannel = net.fc.in_features
net.fc = nn.Linear(inchannel, 5)#重新確定全連接層
完整代碼：
model部分：
import torch.nn as nn
import torch
class BasicBlock(nn.Module):#對(duì)應(yīng)18層和34層所對(duì)應(yīng)的殘差結(jié)構(gòu)（既要有實(shí)線殘差結(jié)構(gòu)功能，也要有虛線殘差結(jié)構(gòu)功能）
expansion = 1#殘差結(jié)構(gòu)主分支上的三個(gè)卷積層是否相同，相同為1，第三層是一二層四倍則為4
def __init__(self, in_channel, out_channel, stride=1, downsample=None):#downsample代表虛線殘差結(jié)構(gòu)選項(xiàng)
super(BasicBlock, self).__init__()
self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,
kernel_size=3, stride=stride, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(out_channel)
self.relu = nn.ReLU()
self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel,
kernel_size=3, stride=1, padding=1, bias=False)
self.bn2 = nn.BatchNorm2d(out_channel)
self.downsample = downsample
def forward(self, x):
identity = x
if self.downsample is not None:
identity = self.downsample(x)#得到捷徑分支的輸出
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out += identity
out = self.relu(out)
return out#得到殘差結(jié)構(gòu)的最終輸出
class Bottleneck(nn.Module):#對(duì)應(yīng)50層、101層和152層所對(duì)應(yīng)的殘差結(jié)構(gòu)
expansion = 4#第三層卷積核個(gè)數(shù)是第一層和第二層的四倍
def __init__(self, in_channel, out_channel, stride=1, downsample=None):
super(Bottleneck, self).__init__()
self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,
kernel_size=1, stride=1, bias=False)
self.bn1 = nn.BatchNorm2d(out_channel)
self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel,
kernel_size=3, stride=stride, bias=False, padding=1)
self.bn2 = nn.BatchNorm2d(out_channel)
self.conv3 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel*self.expansion,
kernel_size=1, stride=1, bias=False)
self.bn3 = nn.BatchNorm2d(out_channel*self.expansion)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
def forward(self, x):
identity = x
if self.downsample is not None:
identity = self.downsample(x)
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv3(out)
out = self.bn3(out)
out += identity
out = self.relu(out)
return out
class ResNet(nn.Module):#定義整個(gè)網(wǎng)絡(luò)的框架部分
#blocks_num是殘差結(jié)構(gòu)的數(shù)目，是一個(gè)列表參數(shù)，block對(duì)應(yīng)哪個(gè)殘差模塊
def __init__(self, block, blocks_num, num_classes=1000, include_top=True):
super(ResNet, self).__init__()
self.include_top = include_top
self.in_channel = 64#通過(guò)第一個(gè)池化層后所得到的特征矩陣的深度
self.conv1 = nn.Conv2d(3, self.in_channel, kernel_size=7, stride=2,
padding=3, bias=False)
self.bn1 = nn.BatchNorm2d(self.in_channel)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, blocks_num[0])
self.layer2 = self._make_layer(block, 128, blocks_num[1], stride=2)
self.layer3 = self._make_layer(block, 256, blocks_num[2], stride=2)
self.layer4 = self._make_layer(block, 512, blocks_num[3], stride=2)
if self.include_top:
self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) # output size = (1, 1)
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
def _make_layer(self, block, channel, block_num, stride=1):#channel：殘差結(jié)構(gòu)中，第一個(gè)卷積層所使用的卷積核的個(gè)數(shù)
downsample = None
if stride != 1 or self.in_channel != channel * block.expansion:#18層和34層會(huì)直接跳過(guò)這個(gè)if語(yǔ)句
downsample = nn.Sequential(
nn.Conv2d(self.in_channel, channel * block.expansion, kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(channel * block.expansion))
layers = []
layers.append(block(self.in_channel, channel, downsample=downsample, stride=stride))
self.in_channel = channel * block.expansion
for _ in range(1, block_num):
layers.append(block(self.in_channel, channel))
return nn.Sequential(*layers)
def forward(self, x):
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
if self.include_top:#默認(rèn)是true
x = self.avgpool(x)
x = torch.flatten(x, 1)
x = self.fc(x)
return x
def resnet34(num_classes=1000, include_top=True):
return ResNet(BasicBlock, [3, 4, 6, 3], num_classes=num_classes, include_top=include_top)
def resnet101(num_classes=1000, include_top=True):
return ResNet(Bottleneck, [3, 4, 23, 3], num_classes=num_classes, include_top=include_top)
訓(xùn)練部分：
import torch
import torch.nn as nn
from torchvision import transforms, datasets
import json
import matplotlib.pyplot as plt
import os
import torch.optim as optim
from model import resnet34, resnet101
import torchvision.models.resnet#ctrl+鼠標(biāo)左鍵點(diǎn)擊即可下載權(quán)重
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print(device)
data_transform = {
"train": transforms.Compose([transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),#和官網(wǎng)初始化方法保持一致
"val": transforms.Compose([transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])}
data_root = os.path.abspath(os.path.join(os.getcwd(), "../..")) # get data root path
image_path = data_root + "/data_set/flower_data/" # flower data set path
train_dataset = datasets.ImageFolder(root=image_path+"train",
transform=data_transform["train"])
train_num = len(train_dataset)
# {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_dataset.class_to_idx
cla_dict = dict((val, key) for key, val in flower_list.items())
# write dict into json file
json_str = json.dumps(cla_dict, indent=4)
with open('class_indices.json', 'w') as json_file:
json_file.write(json_str)
batch_size = 16
train_loader = torch.utils.data.DataLoader(train_dataset,
batch_size=batch_size, shuffle=True,
num_workers=0)
validate_dataset = datasets.ImageFolder(root=image_path + "val",
transform=data_transform["val"])
val_num = len(validate_dataset)
validate_loader = torch.utils.data.DataLoader(validate_dataset,
batch_size=batch_size, shuffle=False,
num_workers=0)
net = resnet34()#一開始不能設(shè)置全連接層的輸出種類為自己想要的，必須先將模型參數(shù)載入，再修改全連接層
# 官方提供載入預(yù)訓(xùn)練模型的方法
model_weight_path = "./resnet34-pre.pth"#權(quán)重路徑
missing_keys, unexpected_keys = net.load_state_dict(torch.load(model_weight_path), strict=False)#載入模型權(quán)重
inchannel = net.fc.in_features
net.fc = nn.Linear(inchannel, 5)#重新確定全連接層
net.to(device)
loss_function = nn.CrossEntropyLoss()
optimizer = optim.Adam(net.parameters(), lr=0.0001)
best_acc = 0.0
save_path = './resNet34.pth'
for epoch in range(3):
# train
net.train()#控制BN層狀態(tài)
running_loss = 0.0
for step, data in enumerate(train_loader, start=0):
images, labels = data
optimizer.zero_grad()
logits = net(images.to(device))
loss = loss_function(logits, labels.to(device))
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
# print train process
rate = (step+1)/len(train_loader)
a = "*" * int(rate * 50)
b = "." * int((1 - rate) * 50)
print("\rtrain loss: {:^3.0f}%[{}->{}]{:.4f}".format(int(rate*100), a, b, loss), end="")
print()
# validate
net.eval()#控制BN層狀態(tài)
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad():
for val_data in validate_loader:
val_images, val_labels = val_data
outputs = net(val_images.to(device)) # eval model only have last output layer
# loss = loss_function(outputs, test_labels)
predict_y = torch.max(outputs, dim=1)[1]
acc += (predict_y == val_labels.to(device)).sum().item()
val_accurate = acc / val_num
if val_accurate > best_acc:
best_acc = val_accurate
torch.save(net.state_dict(), save_path)
print('[epoch %d] train_loss: %.3f test_accuracy: %.3f' %
(epoch + 1, running_loss / step, val_accurate))
print('Finished Training')
預(yù)測(cè)部分：
import torch
from model import resnet34
from PIL import Image
from torchvision import transforms
import matplotlib.pyplot as plt
import json
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
data_transform = transforms.Compose(
[transforms.Resize(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])#采用和訓(xùn)練方法一樣的標(biāo)準(zhǔn)化處理
# load image
img = Image.open("../aa.jpg")
plt.imshow(img)
# [N, C, H, W]
img = data_transform(img)
# expand batch dimension
img = torch.unsqueeze(img, dim=0)
# read class_indict
try:
json_file = open('./class_indices.json', 'r')
class_indict = json.load(json_file)
except Exception as e:
print(e)
exit(-1)
# create model
model = resnet34(num_classes=5)
# load model weights
model_weight_path = "./resNet34.pth"
model.load_state_dict(torch.load(model_weight_path, map_location=device))#載入訓(xùn)練好的模型參數(shù)
model.eval()#使用eval()模式
with torch.no_grad():#不跟蹤損失梯度
# predict class
output = torch.squeeze(model(img))#壓縮batch維度
predict = torch.softmax(output, dim=0)#通過(guò)softmax得到概率分布
predict_cla = torch.argmax(predict).numpy()#尋找最大值所對(duì)應(yīng)的索引
print(class_indict[str(predict_cla)], predict[predict_cla].numpy())#打印類別信息和概率
plt.show()
以上為個(gè)人經(jīng)驗(yàn)，希望能給大家一個(gè)參考，也希望大家多多支持腳本之家。
您可能感興趣的文章:- PyTorch加載預(yù)訓(xùn)練模型實(shí)例(pretrained)
- 獲取Pytorch中間某一層權(quán)重或者特征的例子
- Pytorch修改ResNet模型全連接層進(jìn)行直接訓(xùn)練實(shí)例
