测试三个模型

Author: ZhangXinNan

pytorch-06/README.md

@@ -1,5 +1,11 @@
 
 ## 1 使用net_cnn.py
+```bash
+python cifar.py --model cnn --gpu_id 1
+cnn : val_acc : 0.6813, val_loss : 0.3075759708881378
+```
+
+
 ```bash
 2021-08-07 03:17:48 [10,  2000] train acc: 0.889 loss: 0.308 ; val acc: 0.694 loss: 0.299  lr: 0.001
 2021-08-07 03:18:07 [10,  4000] train acc: 0.869 loss: 0.356 ; val acc: 0.689 loss: 0.295  lr: 0.001
@@ -21,6 +27,11 @@ Accuracy of truck : 68 %
 ```
 
 ## 2 使用net_gap.py
+```bash
+python cifar.py --model gap --gpu_id 1
+gap : val_acc : 0.6217, val_loss : 0.27244704961776733
+```
+
 ```bash
 2021-08-07 03:19:08 [10,  2000] train acc: 0.651 loss: 1.004 ; val acc: 0.616 loss: 0.272  lr: 0.001
 2021-08-07 03:19:27 [10,  4000] train acc: 0.655 loss: 0.982 ; val acc: 0.650 loss: 0.253  lr: 0.001
@@ -43,6 +54,12 @@ Accuracy of truck : 90 %
 
 
 ## 3 net_vgg.py
+```bash
+python cifar.py --model vgg --gpu_id 1
+vgg : val_acc : 0.8284, val_loss : 0.1539972424507141
+```
+
+
 ```bash
 2021-08-07 04:01:42 [10,  2000] train acc: 0.948 loss: 0.156 ; val acc: 0.828 loss: 0.159  lr: 0.001
 2021-08-07 04:02:41 [10,  4000] train acc: 0.946 loss: 0.158 ; val acc: 0.828 loss: 0.158  lr: 0.001

pytorch-06/cifar.py

@@ -51,10 +51,7 @@ def val(net, testloader, criterion):
     return correct / total, loss / total
 
 
-def train(net, trainloader, testloader, num_epoch):
-    criterion = nn.CrossEntropyLoss()
-    optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
-
+def train(net, trainloader, testloader, num_epoch, optimizer, criterion):
     # Print optimizer's state_dict
     print("Optimizer's state_dict:")
     for var_name in optimizer.state_dict():
@@ -126,15 +123,20 @@ def main(args):
     # print(device)
     net.to(device)
 
+    criterion = nn.CrossEntropyLoss()
+    optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
+
     trainloader, testloader = load_data_cifar()
     weights_file = './cifar_net_{}.pth'.format(args.model)
     if not os.path.isfile(weights_file):
-        train(net, trainloader, testloader, args.num_epoch)
+        train(net, trainloader, testloader, args.num_epoch, optimizer, criterion)
         # PATH = './cifar_net.pth'
         torch.save(net.state_dict(), weights_file)
     else:
         net.load_state_dict(torch.load(weights_file))
         net.to(device)
+        val_acc, val_loss = val(net, testloader, criterion)
+        print('{} : val_acc : {}, val_loss : {}'.format(args.model, val_acc, val_loss))
     # 打印每个类别的准确率
     print_accuracy_of_classes(net, testloader)