基于pytorch實現(xiàn)運動鞋品牌識別功能
一、前期準備
1.設置GPU
import torch import torch.nn as nn import torchvision.transforms as transforms import torchvision from torchvision import transforms, datasets import os,PIL,pathlib,random device = torch.device("cuda" if torch.cuda.is_available() else "cpu") device
2. 導入數(shù)據(jù)
data_dir = '../Data/運動鞋品牌識別數(shù)據(jù)/' data_dir = pathlib.Path(data_dir) data_paths = list(data_dir.glob('*/*')) classeNames = sorted(item.name for item in data_dir.glob('*/') if item.is_dir()) classeNames
['test', 'train']
- 第一步:使用
pathlib.Path()
函數(shù)將字符串類型的文件夾路徑轉(zhuǎn)換為pathlib.Path
對象。 - 第二步:使用
glob()
方法獲取data_dir路徑下的所有文件路徑,并以列表形式存儲在data_paths
中。 - 第三步:通過
split()
函數(shù)對data_paths
中的每個文件路徑執(zhí)行分割操作,獲得各個文件所屬的類別名稱,并存儲在classeNames
中 - 第四步:打印
classeNames
列表,顯示每個文件所屬的類別名稱。
train_transforms = transforms.Compose([ transforms.Resize([224, 224]), # 將輸入圖片resize成統(tǒng)一尺寸 # transforms.RandomHorizontalFlip(), # 隨機水平翻轉(zhuǎn) transforms.ToTensor(), # 將PIL Image或numpy.ndarray轉(zhuǎn)換為tensor,并歸一化到[0,1]之間 transforms.Normalize( # 標準化處理-->轉(zhuǎn)換為標準正太分布(高斯分布),使模型更容易收斂 mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]與std=[0.229,0.224,0.225] 從數(shù)據(jù)集中隨機抽樣計算得到的。 ]) test_transform = transforms.Compose([ transforms.Resize([224, 224]), # 將輸入圖片resize成統(tǒng)一尺寸 transforms.ToTensor(), # 將PIL Image或numpy.ndarray轉(zhuǎn)換為tensor,并歸一化到[0,1]之間 transforms.Normalize( # 標準化處理-->轉(zhuǎn)換為標準正太分布(高斯分布),使模型更容易收斂 mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) # 其中 mean=[0.485,0.456,0.406]與std=[0.229,0.224,0.225] 從數(shù)據(jù)集中隨機抽樣計算得到的。 ]) train_dataset = datasets.ImageFolder("../Data/運動鞋品牌識別數(shù)據(jù)/train/",transform=train_transforms) test_dataset = datasets.ImageFolder("../Data/運動鞋品牌識別數(shù)據(jù)/test/",transform=train_transforms)
train_dataset.class_to_idx
{'adidas': 0, 'nike': 1}
3. 劃分數(shù)據(jù)集
batch_size = 32 train_dl = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=1) test_dl = torch.utils.data.DataLoader(test_dataset, batch_size=batch_size, shuffle=True, num_workers=1)
for X, y in test_dl: print("Shape of X [N, C, H, W]: ", X.shape) print("Shape of y: ", y.shape, y.dtype) break
Shape of X [N, C, H, W]: torch.Size([32, 3, 224, 224]) Shape of y: torch.Size([32]) torch.int64
二、構(gòu)建簡單的CNN網(wǎng)絡
對于一般的CNN網(wǎng)絡來說,都是由特征提取網(wǎng)絡和分類網(wǎng)絡構(gòu)成,其中特征提取網(wǎng)絡用于提取圖片的特征,分類網(wǎng)絡用于將圖片進行分類。
網(wǎng)絡結(jié)構(gòu)圖
import torch.nn.functional as F class Model(nn.Module): def __init__(self): super(Model, self).__init__() self.conv1=nn.Sequential( nn.Conv2d(3, 12, kernel_size=5, padding=0), # 12*220*220 nn.BatchNorm2d(12), nn.ReLU()) self.conv2=nn.Sequential( nn.Conv2d(12, 12, kernel_size=5, padding=0), # 12*216*216 nn.BatchNorm2d(12), nn.ReLU()) self.pool3=nn.Sequential( nn.MaxPool2d(2)) # 12*108*108 self.conv4=nn.Sequential( nn.Conv2d(12, 24, kernel_size=5, padding=0), # 24*104*104 nn.BatchNorm2d(24), nn.ReLU()) self.conv5=nn.Sequential( nn.Conv2d(24, 24, kernel_size=5, padding=0), # 24*100*100 nn.BatchNorm2d(24), nn.ReLU()) self.pool6=nn.Sequential( nn.MaxPool2d(2)) # 24*50*50 self.dropout = nn.Sequential( nn.Dropout(0.2)) self.fc=nn.Sequential( nn.Linear(24*50*50, len(classeNames))) def forward(self, x): batch_size = x.size(0) x = self.conv1(x) # 卷積-BN-激活 x = self.conv2(x) # 卷積-BN-激活 x = self.pool3(x) # 池化 x = self.conv4(x) # 卷積-BN-激活 x = self.conv5(x) # 卷積-BN-激活 x = self.pool6(x) # 池化 x = self.dropout(x) x = x.view(batch_size, -1) # flatten 變成全連接網(wǎng)絡需要的輸入 (batch, 24*50*50) ==> (batch, -1), -1 此處自動算出的是24*50*50 x = self.fc(x) return x device = "cuda" if torch.cuda.is_available() else "cpu" print("Using {} device".format(device)) model = Model().to(device) model
Model( (conv1): Sequential( (0): Conv2d(3, 12, kernel_size=(5, 5), stride=(1, 1)) (1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (2): ReLU() ) (conv2): Sequential( (0): Conv2d(12, 12, kernel_size=(5, 5), stride=(1, 1)) (1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (2): ReLU() ) (pool3): Sequential( (0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) ) (conv4): Sequential( (0): Conv2d(12, 24, kernel_size=(5, 5), stride=(1, 1)) (1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (2): ReLU() ) (conv5): Sequential( (0): Conv2d(24, 24, kernel_size=(5, 5), stride=(1, 1)) (1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (2): ReLU() ) (pool6): Sequential( (0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) ) (dropout): Sequential( (0): Dropout(p=0.2, inplace=False) ) (fc): Sequential( (0): Linear(in_features=60000, out_features=2, bias=True) ) )
三、 訓練模型
1. 編寫訓練函數(shù)
# 訓練循環(huán) def train(dataloader, model, loss_fn, optimizer): size = len(dataloader.dataset) # 訓練集的大小 num_batches = len(dataloader) # 批次數(shù)目, (size/batch_size,向上取整) train_loss, train_acc = 0, 0 # 初始化訓練損失和正確率 for X, y in dataloader: # 獲取圖片及其標簽 X, y = X.to(device), y.to(device) # 計算預測誤差 pred = model(X) # 網(wǎng)絡輸出 loss = loss_fn(pred, y) # 計算網(wǎng)絡輸出和真實值之間的差距,targets為真實值,計算二者差值即為損失 # 反向傳播 optimizer.zero_grad() # grad屬性歸零 loss.backward() # 反向傳播 optimizer.step() # 每一步自動更新 # 記錄acc與loss train_acc += (pred.argmax(1) == y).type(torch.float).sum().item() train_loss += loss.item() train_acc /= size train_loss /= num_batches return train_acc, train_loss
2. 編寫測試函數(shù)
def test (dataloader, model, loss_fn): size = len(dataloader.dataset) # 測試集的大小 num_batches = len(dataloader) # 批次數(shù)目, (size/batch_size,向上取整) test_loss, test_acc = 0, 0 # 當不進行訓練時,停止梯度更新,節(jié)省計算內(nèi)存消耗 with torch.no_grad(): for imgs, target in dataloader: imgs, target = imgs.to(device), target.to(device) # 計算loss target_pred = model(imgs) loss = loss_fn(target_pred, target) test_loss += loss.item() test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item() test_acc /= size test_loss /= num_batches return test_acc, test_loss
3. 設置動態(tài)學習率
def adjust_learning_rate(optimizer, epoch, start_lr): # 每 2 個epoch衰減到原來的 0.98 lr = start_lr * (0.92 ** (epoch // 2)) for param_group in optimizer.param_groups: param_group['lr'] = lr learn_rate = 1e-4 # 初始學習率 optimizer = torch.optim.SGD(model.parameters(), lr=learn_rate)
4. 正式訓練
loss_fn = nn.CrossEntropyLoss() # 創(chuàng)建損失函數(shù) epochs = 40 train_loss = [] train_acc = [] test_loss = [] test_acc = [] for epoch in range(epochs): # 更新學習率(使用自定義學習率時使用) adjust_learning_rate(optimizer, epoch, learn_rate) model.train() epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer) # scheduler.step() # 更新學習率(調(diào)用官方動態(tài)學習率接口時使用) model.eval() epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn) train_acc.append(epoch_train_acc) train_loss.append(epoch_train_loss) test_acc.append(epoch_test_acc) test_loss.append(epoch_test_loss) # 獲取當前的學習率 lr = optimizer.state_dict()['param_groups'][0]['lr'] template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}') print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss, epoch_test_acc*100, epoch_test_loss, lr)) print('Done')
Epoch: 1, Train_acc:52.6%, Train_loss:0.744, Test_acc:50.0%, Test_loss:0.716, Lr:1.00E-04 Epoch: 2, Train_acc:59.0%, Train_loss:0.690, Test_acc:67.1%, Test_loss:0.618, Lr:1.00E-04 Epoch: 3, Train_acc:64.1%, Train_loss:0.627, Test_acc:61.8%, Test_loss:0.637, Lr:9.20E-05 Epoch: 4, Train_acc:67.9%, Train_loss:0.588, Test_acc:77.6%, Test_loss:0.584, Lr:9.20E-05 Epoch: 5, Train_acc:74.7%, Train_loss:0.539, Test_acc:73.7%, Test_loss:0.553, Lr:8.46E-05 Epoch: 6, Train_acc:76.3%, Train_loss:0.516, Test_acc:76.3%, Test_loss:0.528, Lr:8.46E-05 Epoch: 7, Train_acc:77.1%, Train_loss:0.495, Test_acc:80.3%, Test_loss:0.533, Lr:7.79E-05 Epoch: 8, Train_acc:77.3%, Train_loss:0.491, Test_acc:76.3%, Test_loss:0.548, Lr:7.79E-05 Epoch: 9, Train_acc:78.1%, Train_loss:0.457, Test_acc:76.3%, Test_loss:0.516, Lr:7.16E-05 Epoch:10, Train_acc:83.1%, Train_loss:0.436, Test_acc:73.7%, Test_loss:0.513, Lr:7.16E-05 Epoch:11, Train_acc:81.5%, Train_loss:0.442, Test_acc:77.6%, Test_loss:0.525, Lr:6.59E-05 Epoch:12, Train_acc:83.3%, Train_loss:0.423, Test_acc:75.0%, Test_loss:0.552, Lr:6.59E-05 Epoch:13, Train_acc:82.3%, Train_loss:0.418, Test_acc:77.6%, Test_loss:0.477, Lr:6.06E-05 Epoch:14, Train_acc:85.3%, Train_loss:0.403, Test_acc:76.3%, Test_loss:0.513, Lr:6.06E-05 Epoch:15, Train_acc:86.1%, Train_loss:0.387, Test_acc:78.9%, Test_loss:0.509, Lr:5.58E-05 Epoch:16, Train_acc:87.5%, Train_loss:0.372, Test_acc:80.3%, Test_loss:0.486, Lr:5.58E-05 Epoch:17, Train_acc:88.2%, Train_loss:0.358, Test_acc:75.0%, Test_loss:0.460, Lr:5.13E-05 Epoch:18, Train_acc:88.2%, Train_loss:0.359, Test_acc:77.6%, Test_loss:0.469, Lr:5.13E-05 Epoch:19, Train_acc:88.6%, Train_loss:0.360, Test_acc:78.9%, Test_loss:0.504, Lr:4.72E-05 Epoch:20, Train_acc:89.4%, Train_loss:0.357, Test_acc:78.9%, Test_loss:0.480, Lr:4.72E-05 Epoch:21, Train_acc:90.4%, Train_loss:0.341, Test_acc:78.9%, Test_loss:0.475, Lr:4.34E-05 Epoch:22, Train_acc:90.2%, Train_loss:0.335, Test_acc:78.9%, Test_loss:0.481, Lr:4.34E-05 Epoch:23, Train_acc:89.4%, Train_loss:0.335, Test_acc:77.6%, Test_loss:0.491, Lr:4.00E-05 Epoch:24, Train_acc:91.4%, Train_loss:0.320, Test_acc:78.9%, Test_loss:0.469, Lr:4.00E-05 Epoch:25, Train_acc:92.6%, Train_loss:0.324, Test_acc:78.9%, Test_loss:0.485, Lr:3.68E-05 Epoch:26, Train_acc:92.4%, Train_loss:0.313, Test_acc:78.9%, Test_loss:0.478, Lr:3.68E-05 Epoch:27, Train_acc:91.8%, Train_loss:0.307, Test_acc:77.6%, Test_loss:0.436, Lr:3.38E-05 Epoch:28, Train_acc:90.4%, Train_loss:0.313, Test_acc:77.6%, Test_loss:0.480, Lr:3.38E-05 Epoch:29, Train_acc:93.0%, Train_loss:0.302, Test_acc:76.3%, Test_loss:0.485, Lr:3.11E-05 Epoch:30, Train_acc:92.2%, Train_loss:0.306, Test_acc:78.9%, Test_loss:0.438, Lr:3.11E-05 Epoch:31, Train_acc:92.4%, Train_loss:0.306, Test_acc:77.6%, Test_loss:0.455, Lr:2.86E-05 Epoch:32, Train_acc:92.6%, Train_loss:0.299, Test_acc:78.9%, Test_loss:0.425, Lr:2.86E-05 Epoch:33, Train_acc:91.6%, Train_loss:0.299, Test_acc:77.6%, Test_loss:0.524, Lr:2.63E-05 Epoch:34, Train_acc:93.6%, Train_loss:0.290, Test_acc:78.9%, Test_loss:0.477, Lr:2.63E-05 Epoch:35, Train_acc:94.0%, Train_loss:0.290, Test_acc:78.9%, Test_loss:0.455, Lr:2.42E-05 Epoch:36, Train_acc:93.4%, Train_loss:0.282, Test_acc:78.9%, Test_loss:0.453, Lr:2.42E-05 Epoch:37, Train_acc:94.2%, Train_loss:0.281, Test_acc:78.9%, Test_loss:0.457, Lr:2.23E-05 Epoch:38, Train_acc:94.0%, Train_loss:0.289, Test_acc:78.9%, Test_loss:0.449, Lr:2.23E-05 Epoch:39, Train_acc:94.2%, Train_loss:0.279, Test_acc:77.6%, Test_loss:0.435, Lr:2.05E-05 Epoch:40, Train_acc:93.8%, Train_loss:0.280, Test_acc:77.6%, Test_loss:0.425, Lr:2.05E-05 Done
四、 結(jié)果可視化
1. Loss與Accuracy圖
import matplotlib.pyplot as plt #隱藏警告 import warnings warnings.filterwarnings("ignore") #忽略警告信息 plt.rcParams['font.sans-serif'] = ['SimHei'] # 用來正常顯示中文標簽 plt.rcParams['axes.unicode_minus'] = False # 用來正常顯示負號 plt.rcParams['figure.dpi'] = 100 #分辨率 epochs_range = range(epochs) plt.figure(figsize=(12, 3)) plt.subplot(1, 2, 1) plt.plot(epochs_range, train_acc, label='Training Accuracy') plt.plot(epochs_range, test_acc, label='Test Accuracy') plt.legend(loc='lower right') plt.title('Training and Validation Accuracy') plt.subplot(1, 2, 2) plt.plot(epochs_range, train_loss, label='Training Loss') plt.plot(epochs_range, test_loss, label='Test Loss') plt.legend(loc='upper right') plt.title('Training and Validation Loss') plt.show()
2. 指定圖片進行預測
from PIL import Image classes = list(train_dataset.class_to_idx) def predict_one_image(image_path, model, transform, classes): test_img = Image.open(image_path).convert('RGB') # plt.imshow(test_img) # 展示預測的圖片 test_img = transform(test_img) img = test_img.to(device).unsqueeze(0) model.eval() output = model(img) _,pred = torch.max(output,1) pred_class = classes[pred] print(f'預測結(jié)果是:{pred_class}')
# 預測訓練集中的某張照片 predict_one_image(image_path='../Data/運動鞋品牌識別數(shù)據(jù)/test/nike/31.jpg', model=model, transform=train_transforms, classes=classes)
預測結(jié)果是:nike
五、保存并加載模型
# 模型保存 PATH = './model/shoeBrands_model.pth' # 保存的參數(shù)文件名 torch.save(model.state_dict(), PATH) # 將參數(shù)加載到model當中 model.load_state_dict(torch.load(PATH, map_location=device))
六、動態(tài)學習率
1.torch.optim.lr_scheduler.StepLR
等間隔動態(tài)調(diào)整方法,每經(jīng)過step_size個epoch,做一次學習率decay,以gamma值為縮小倍數(shù)。
函數(shù)原型:
torch.optim.lr_scheduler.StepLR(optimizer, step_size, gamma=0.1, last_epoch=-1)
參數(shù):
- optimizer(Optimizer):要調(diào)整學習率的優(yōu)化器
- step_size(int):學習率調(diào)整的間隔epoch數(shù)
- gamma(float):學習率調(diào)整的縮減比例
- last_epoch(int):上一次調(diào)整學習率的時間點,默認為-1
代碼示例:
optimizer = torch.optim.SGD(net.parameters(), lr=0.001 )
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.1)
2.torch.optim.lr_scheduler.LambdaLR
根據(jù)給定的函數(shù)動態(tài)調(diào)整學習率。
函數(shù)原型:
torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda, last_epoch=-1)
參數(shù):
- optimizer(Optimizer):要調(diào)整學習率的優(yōu)化器
- lr_lambda(function):根據(jù)epoch返回一個值,作為學習率的倍數(shù)
- last_epoch(int):上一次調(diào)整學習率的時間點,默認為-1
代碼示例:
lambda1 = lambda epoch: (0.92 ** (epoch // 2) # 第二組參數(shù)的調(diào)整方法
optimizer = torch.optim.SGD(model.parameters(), lr=learn_rate)
scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lambda1) #選定調(diào)整方法
3.torch.optim.lr_scheduler.MultiStepLR
等間隔動態(tài)調(diào)整方法,在指定的epoch位置做一次學習率decay,以gamma值為縮小倍數(shù)。
函數(shù)原型:
torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones, gamma=0.1, last_epoch=-1)
參數(shù):
- optimizer(Optimizer):要調(diào)整學習率的優(yōu)化器
- milestones(list):學習率調(diào)整的epoch位置
- gamma(float):學習率調(diào)整的縮減比例
- last_epoch(int):上一次調(diào)整學習率的時間點,默認為-1
代碼示例:
optimizer = torch.optim.SGD(net.parameters(), lr=0.001 )
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,milestones=[2,6,15],gamma=0.1)
七、個人收獲
在這個項目中,我首先準備了數(shù)據(jù),包括設置GPU環(huán)境、導入數(shù)據(jù)、劃分數(shù)據(jù)集等。然后構(gòu)建了一個簡單的CNN網(wǎng)絡,用于對運動鞋品牌進行識別。接著,我編寫了訓練函數(shù)和測試函數(shù),用于訓練模型和評估模型性能。在訓練過程中,我還使用了動態(tài)學習率的方法,通過調(diào)整學習率來優(yōu)化模型訓練過程。最后,我展示了訓練過程中的損失和準確率的變化情況,并對模型進行了保存和加載,以便后續(xù)的使用。
通過這個項目,我深入了解了深度學習模型的訓練流程,包括數(shù)據(jù)準備、模型構(gòu)建、訓練和評估,以及模型的保存和加載。同時,動態(tài)學習率的應用也豐富了我的訓練優(yōu)化方法的知識儲備。這些知識將對我未來的深度學習項目產(chǎn)生積極的影響。
以上就是基于pytorch實現(xiàn)運動鞋品牌識別功能的詳細內(nèi)容,更多關于pytorch運動鞋品牌識別的資料請關注腳本之家其它相關文章!
相關文章
Python lambda和Python def區(qū)別分析
Python支持一種有趣的語法,它允許你快速定義單行的最小函數(shù)。這些叫做lambda的函數(shù),是從Lisp借用來的,可以用在任何需要函數(shù)的地方2014-11-11聊聊prod()與cumprod()區(qū)別cumsum()
這篇文章主要介紹了prod()與cumprod()區(qū)別cumsum(),具有很好的參考價值,希望對大家有所幫助。如有錯誤或未考慮完全的地方,望不吝賜教2021-05-05windows10 pycharm下安裝pyltp庫和加載模型實現(xiàn)語義角色標注的示例代碼
這篇文章主要介紹了windows10 pycharm下安裝pyltp庫和加載模型實現(xiàn)語義角色標注,本文通過圖文實例相結(jié)合給大家介紹的非常詳細,對大家的學習或工作具有一定的參考借鑒價值,需要的朋友可以參考下2020-05-05一文深入學習Python中的os.listdir函數(shù)
這篇文章主要給大家介紹了關于Python中os.listdir函數(shù)的相關資料,os.listdir是 Python中的一個函數(shù),它的意思是返回指定目錄下的文件和文件夾的名稱的列表,需要的朋友可以參考下2023-10-10pandas 實現(xiàn)將NaN轉(zhuǎn)換為None
這篇文章主要介紹了pandas 實現(xiàn)將NaN轉(zhuǎn)換為None的操作,具有很好的參考價值,希望對大家有所幫助。如有錯誤或未考慮完全的地方,望不吝賜教2021-05-05