diff --git a/src/veggies_recognition/best_model.pth b/src/veggies_recognition/best_model.pth
new file mode 100644
index 00000000..6fe3a9a2
Binary files /dev/null and b/src/veggies_recognition/best_model.pth differ
diff --git a/src/veggies_recognition/model_best_checkpoint.pth.tar b/src/veggies_recognition/model_best_checkpoint.pth.tar
new file mode 100644
index 00000000..11236dcd
Binary files /dev/null and b/src/veggies_recognition/model_best_checkpoint.pth.tar differ
diff --git a/src/veggies_recognition/train.py b/src/veggies_recognition/train.py
index da3b1179..849d4b88 100644
--- a/src/veggies_recognition/train.py
+++ b/src/veggies_recognition/train.py
@@ -37,6 +37,8 @@ device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 print(device)
 
 def train_nn(model, train_loader, test_loader, criterion, optimizer, num_epochs):
+    best_acc = 0
+
     for epoch in range(num_epochs):
         print('Epoch number %d' % (epoch+1))
         model.train()
@@ -64,10 +66,24 @@ def train_nn(model, train_loader, test_loader, criterion, optimizer, num_epochs)
 
         print('    ---Training--- Got %d out of %d images correctly (%.3f%%). Epoch loss: %.3f' % (running_correct, total, epoch_acc, epoch_loss))
 
-        evaluate_model_on_test_set(model, test_loader)
+        test_dataset_acc = evaluate_model_on_test_set(model, test_loader)
+
+        if(test_dataset_acc>best_acc):
+            best_acc = test_dataset_acc
+            save_checkpoint(model, epoch, optimizer, best_acc)
+
     print('Finished Training')
     return model
 
+def save_checkpoint(model, epoch, optimizer, best_acc):
+    state = {
+        'epoch': epoch +1,
+        'model': model.state_dict(),
+        'optimizer': optimizer.state_dict(),
+        'best_acc': best_acc
+        }
+    torch.save(state, 'model_best_checkpoint.pth.tar')
+
 def evaluate_model_on_test_set(model, test_loader):
     model.eval()
     predicted_correctly_on_epoch = 0
@@ -84,6 +100,7 @@ def evaluate_model_on_test_set(model, test_loader):
             predicted_correctly_on_epoch += (predicted==labels).sum().item()
     epoch_acc = predicted_correctly_on_epoch / total *100
     print('    ---Testing--- Got %d out of %d images correctly (%.3f%%).' % (predicted_correctly_on_epoch, total, epoch_acc))
+    return epoch_acc
 
 
 resnet18_model = models.resnet18(pretrained=False)
@@ -95,4 +112,14 @@ loss_fn = nn.CrossEntropyLoss()
 
 optimizer = optim.SGD(resnet18_model.parameters(), lr=0.01, momentum=0.9, weight_decay=0.003)
 
-train_nn(resnet18_model, train_loader, test_loader, loss_fn, optimizer, 20)
\ No newline at end of file
+train_nn(resnet18_model, train_loader, test_loader, loss_fn, optimizer, 30)
+
+checkpoint = torch.load('model_best_checkpoint.pth.tar')
+
+resnet18_model = models.resnet18()
+num_features = resnet18_model.fc.in_features
+number_of_classes = 25
+resnet18_model.fc = nn.Linear(num_features, number_of_classes)
+resnet18_model.load_state_dict(checkpoint['model'])
+
+torch.save(resnet18_model, 'best_model.pth')
\ No newline at end of file