1def get_transforms1(*, data):
2
3    #train,valid以外だったら処理を止める。
4    
5    if data == 'train':
6        return Compose([
7            HorizontalFlip(p=0.5),
8            VerticalFlip(p=0.5),
9            GaussNoise(p=0.5),
10            RandomRotate90(p=0.5),
11            RandomGamma(p=0.5),
12            RandomAugMix(severity=3, width=3, alpha=1., p=0.5),
13            Normalize(
14                mean=[0.485, 0.456, 0.406],
15                std=[0.229, 0.224, 0.225],
16            )
17        ])
18    elif data == 'valid':
19        return Compose([
20            Normalize(
21                mean=[0.485, 0.456, 0.406],
22                std=[0.229, 0.224, 0.225],
23            )
24        ])
25    
26def to_tensor(*args):
27
28        return Compose([
29            ToTensor()
30        ])

1class TrainDataset(Dataset):
2    def __init__(self, df, labels, transform1=None, transform2=None):
3        self.df = df
4        self.labels = labels
5        self.transform = transform1
6        self.transform_ = transform2
7        
8    def __len__(self):
9        return len(self.df)
10
11    def __getitem__(self, idx):
12        file_name = self.df['filename'].values[idx]
13        file_path = '/kaggle/input/computed-tomography-of-lungs-datase-for-covid19/{}'.format(file_name)
14        image = cv2.imread(file_path)
15        image = cv2.resize(image,(SIZE,SIZE))
16        if self.transform:
17            image = self.transform(image=image)['image']
18        if self.transform_:
19            image = self.transform_(image=image)['image']
20            
21        label = torch.tensor(self.labels[idx]).float()
22        return image, label

1class Efnet_b2_ns(nn.Module):
2
3    def __init__(self,weight_path):
4        super().__init__()
5        self.weight_path = weight_path
6        self.model = geffnet.tf_efficientnet_b2_ns(pretrained=False)
7        state_dict = torch.load(self.weight_path,map_location=device)
8        self.model.load_state_dict(fix_model_state_dict(state_dict))
9        #さいごの部分を付け替え
10        self.model.global_pool=nn.AdaptiveAvgPool2d(1)
11        self.model.classifier = nn.Linear(self.model.classifier.in_features, 1)
12        
13        
14    def forward(self, x):
15        x = self.model(x)#ベースのモデルの流れに同じ
16        return x
17def fix_model_state_dict(state_dict):
18    from collections import OrderedDict
19    new_state_dict = OrderedDict()
20    for k, v in state_dict.items():
21        name = k
22        if name.startswith('model.'):
23            name = name[6:]  # remove 'model.' of dataparallel
24        new_state_dict[name] = v
25    return new_state_dict