wf c963r代碼1e

前沿拓展：

---

深度學習的比賽中，圖片分類是很常見的比賽，同時也是很難取得特別高名次的比賽，因為圖片分類已經被大家研究的很透徹，一些開源的網絡很容易取得高分。如果大家還掌握不了使用開源的網絡進行訓練，再慢慢去模型調優，很難取得較好的成績。

1.數據介紹

數據下載地址：

https://download.csdn.net/download/xiaosongshine/11128410

這次的實戰使用的數據是交通標志數據集，共有62類交通標志。其中訓練集數據有4572張照片（每個類別大概七十個），測試數據集有2520張照片（每個類別大概40個）。數據包含兩個子目錄分別train與test：

為什么還需要測試數據集呢？這個測試數據集不會拿來訓練，是用來進行模型的評估與調優。

train與test每個文件夾里又有62個子文件夾，每個類別在同一個文件夾內：

我從中打開一個文件間，把里面圖片展示出來：

其中每張照片都類似下面的例子，100*100*3的大小。100是照片的照片的長和寬，3是什么呢？這其實是照片的色彩通道數目，RGB。彩色照片存儲在計算機里就是以三維數組的形式。我們送入網絡的也是這些數組。

2.網絡構建

1.導入Python包，定義一些參數

1import torch as t 2import torchvision as tv 3import os 4import time 5import numpy as np 6from tqdm import tqdm 7 8 9class DefaultConfigs(object): 10 11 data_dir = "./traffic-sign/" 12 data_list = ["train","test"] 13 14 lr = 0.001 15 epochs = 10 16 num_classes = 62 17 image_size = 224 18 batch_size = 40 19 channels = 3 20 gpu = "0" 21 train_len = 4572 22 test_len = 2520 23 use_gpu = t.cuda.is_available() 24 25config = DefaultConfigs()

2.數據準備，采用PyTorch提供的讀取方式

注意一點Train數據需要進行隨機裁剪，Test數據不要進行裁剪了

1normalize = tv.transforms.Normalize(mean = [0.485, 0.456, 0.406], 2 std = [0.229, 0.224, 0.225] 3 ) 4 5transform = { 6 config.data_list[0]:tv.transforms.Compose( 7 [tv.transforms.Resize([224,224]),tv.transforms.CenterCrop([224,224]), 8 tv.transforms.ToTensor(),normalize]#tv.transforms.Resize 用于重設圖片大小 9 ) , 10 config.data_list[1]:tv.transforms.Compose( 11 [tv.transforms.Resize([224,224]),tv.transforms.ToTensor(),normalize] 12 ) 13} 14 15datasets = { 16 x:tv.datasets.ImageFolder(root = os.path.join(config.data_dir,x),transform=transform[x]) 17 for x in config.data_list 18} 19 20dataloader = { 21 x:t.utils.data.DataLoader(dataset= datasets[x], 22 batch_size=config.batch_size, 23 shuffle=True 24 ) 25 for x in config.data_list 26}

3.構建網絡模型（使用resnet18進行遷移學習，訓練參數為最后一個全連接層 t.nn.Linear(512,num_classes)）

1def get_model(num_classes): 2 3 model = tv.models.resnet18(pretrained=True) 4 for parma in model.parameters(): 5 parma.requires_grad = False 6 model.fc = t.nn.Sequential( 7 t.nn.Dropout(p=0.3), 8 t.nn.Linear(512,num_classes) 9 ) 10 return(model)

如果電腦硬件支持，可以把下述代碼屏蔽，則訓練整個網絡，最終準確率會上升，訓練數據會變慢。

1for parma in model.parameters(): 2 parma.requires_grad = False

模型輸出

1ResNet( 2 (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False) 3 (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 4 (relu): ReLU(inplace) 5 (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False) 6 (layer1): Sequential( 7 (0): BasicBlock( 8 (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 9 (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 10 (relu): ReLU(inplace) 11 (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 12 (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 13 ) 14 (1): BasicBlock( 15 (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 16 (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 17 (relu): ReLU(inplace) 18 (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 19 (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 20 ) 21 ) 22 (layer2): Sequential( 23 (0): BasicBlock( 24 (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False) 25 (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 26 (relu): ReLU(inplace) 27 (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 28 (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 29 (downsample): Sequential( 30 (0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False) 31 (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 32 ) 33 ) 34 (1): BasicBlock( 35 (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 36 (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 37 (relu): ReLU(inplace) 38 (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 39 (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 40 ) 41 ) 42 (layer3): Sequential( 43 (0): BasicBlock( 44 (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False) 45 (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 46 (relu): ReLU(inplace) 47 (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 48 (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 49 (downsample): Sequential( 50 (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False) 51 (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 52 ) 53 ) 54 (1): BasicBlock( 55 (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 56 (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 57 (relu): ReLU(inplace) 58 (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 59 (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 60 ) 61 ) 62 (layer4): Sequential( 63 (0): BasicBlock( 64 (conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False) 65 (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 66 (relu): ReLU(inplace) 67 (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 68 (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 69 (downsample): Sequential( 70 (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False) 71 (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 72 ) 73 ) 74 (1): BasicBlock( 75 (conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 76 (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 77 (relu): ReLU(inplace) 78 (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False) 79 (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) 80 ) 81 ) 82 (avgpool): AvgPool2d(kernel_size=7, stride=1, padding=0) 83 (fc): Sequential( 84 (0): Dropout(p=0.3) 85 (1): Linear(in_features=512, out_features=62, bias=True) 86 ) 87)

4.訓練模型（支持自動GPU加速）

1def train(epochs): 2 3 model = get_model(config.num_classes) 4 print(model) 5 loss_f = t.nn.CrossEntropyLoss() 6 if(config.use_gpu): 7 model = model.cuda() 8 loss_f = loss_f.cuda() 9 10 opt = t.optim.Adam(model.fc.parameters(),lr = config.lr) 11 time_start = time.time() 12 13 for epoch in range(epochs): 14 train_loss = [] 15 train_acc = [] 16 test_loss = [] 17 test_acc = [] 18 model.train(True) 19 print("Epoch {}/{}".format(epoch+1,epochs)) 20 for batch, datas in tqdm(enumerate(iter(dataloader["train"]))): 21 x,y = datas 22 if (config.use_gpu): 23 x,y = x.cuda(),y.cuda() 24 y_ = model(x) 25 #print(x.shape,y.shape,y_.shape) 26 _, pre_y_ = t.max(y_,1) 27 pre_y = y 28 #print(y_.shape) 29 loss = loss_f(y_,pre_y) 30 #print(y_.shape) 31 acc = t.sum(pre_y_ == pre_y) 32 33 loss.backward() 34 opt.step() 35 opt.zero_grad() 36 if(config.use_gpu): 37 loss = loss.cpu() 38 acc = acc.cpu() 39 train_loss.append(loss.data) 40 train_acc.append(acc) 41 #if((batch+1)%5 ==0): 42 time_end = time.time() 43 print("Batch {}, Train loss:{:.4f}, Train acc:{:.4f}, Time: {}"\ 44 .format(batch+1,np.mean(train_loss)/config.batch_size,np.mean(train_acc)/config.batch_size,(time_end-time_start))) 45 time_start = time.time() 46 47 model.train(False) 48 for batch, datas in tqdm(enumerate(iter(dataloader["test"]))): 49 x,y = datas 50 if (config.use_gpu): 51 x,y = x.cuda(),y.cuda() 52 y_ = model(x) 53 #print(x.shape,y.shape,y_.shape) 54 _, pre_y_ = t.max(y_,1) 55 pre_y = y 56 #print(y_.shape) 57 loss = loss_f(y_,pre_y) 58 acc = t.sum(pre_y_ == pre_y) 59 60 if(config.use_gpu): 61 loss = loss.cpu() 62 acc = acc.cpu() 63 64 test_loss.append(loss.data) 65 test_acc.append(acc) 66 print("Batch {}, Test loss:{:.4f}, Test acc:{:.4f}".format(batch+1,np.mean(test_loss)/config.batch_size,np.mean(test_acc)/config.batch_size)) 67 68 t.save(model,str(epoch+1)+"ttmodel.pkl") 69 70 71 72if __name__ == "__main__": 73 train(config.epochs)

訓練結果如下：

1Epoch 1/10 2115it [00:48, 2.63it/s] 3Batch 115, Train loss:0.0590, Train acc:0.4635, Time: 48.985504150390625 463it [00:24, 2.62it/s] 5Batch 63, Test loss:0.0374, Test acc:0.6790, Time :24.648272275924683 6Epoch 2/10 7115it [00:45, 3.22it/s] 8Batch 115, Train loss:0.0271, Train acc:0.7576, Time: 45.68823838233948 963it [00:23, 2.62it/s] 10Batch 63, Test loss:0.0255, Test acc:0.7524, Time :23.271782875061035 11Epoch 3/10 12115it [00:45, 3.19it/s] 13Batch 115, Train loss:0.0181, Train acc:0.8300, Time: 45.92648506164551 1463it [00:23, 2.60it/s] 15Batch 63, Test loss:0.0212, Test acc:0.7861, Time :23.80789279937744 16Epoch 4/10 17115it [00:45, 3.28it/s] 18Batch 115, Train loss:0.0138, Train acc:0.8767, Time: 45.27525019645691 1963it [00:23, 2.57it/s] 20Batch 63, Test loss:0.0173, Test acc:0.8385, Time :23.736321449279785 21Epoch 5/10 22115it [00:44, 3.22it/s] 23Batch 115, Train loss:0.0112, Train acc:0.8950, Time: 44.983638286590576 2463it [00:22, 2.69it/s] 25Batch 63, Test loss:0.0156, Test acc:0.8520, Time :22.790074348449707 26Epoch 6/10 27115it [00:44, 3.19it/s] 28Batch 115, Train loss:0.0095, Train acc:0.9159, Time: 45.10426950454712 2963it [00:22, 2.77it/s] 30Batch 63, Test loss:0.0158, Test acc:0.8214, Time :22.80412459373474 31Epoch 7/10 32115it [00:45, 2.95it/s] 33Batch 115, Train loss:0.0081, Train acc:0.9280, Time: 45.30439043045044 3463it [00:23, 2.66it/s] 35Batch 63, Test loss:0.0139, Test acc:0.8528, Time :23.122379541397095 36Epoch 8/10 37115it [00:44, 3.23it/s] 38Batch 115, Train loss:0.0073, Train acc:0.9300, Time: 44.304762840270996 3963it [00:22, 2.74it/s] 40Batch 63, Test loss:0.0142, Test acc:0.8496, Time :22.801835536956787 41Epoch 9/10 42115it [00:43, 3.19it/s] 43Batch 115, Train loss:0.0068, Train acc:0.9361, Time: 44.08414030075073 4463it [00:23, 2.44it/s] 45Batch 63, Test loss:0.0142, Test acc:0.8437, Time :23.604419231414795 46Epoch 10/10 47115it [00:46, 3.12it/s] 48Batch 115, Train loss:0.0063, Train acc:0.9337, Time: 46.76597046852112 4963it [00:24, 2.65it/s] 50Batch 63, Test loss:0.0130, Test acc:0.8591, Time :24.64351773262024

訓練10個Epoch，測試集準確率可以到達0.86，已經達到不錯效果。通過修改參數，增加訓練，可以達到更高的準確率。

想要獲取更多人工智能方面的資料

可以加V、、信：hcgx0904(備注“人工智能”）

點擊了解更多，

開始學習《深度學習&計算機視覺精講》。

拓展知識：