level14taken

class ChannelAttentionGate(nn.Module):
    def __init__(self, channel, reduction=16):
        super(ChannelAttentionGate, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
                nn.Linear(channel, channel // reduction),
                nn.ReLU(inplace=True),
                nn.Linear(channel // reduction, channel),
                nn.Sigmoid()
                 )

level14taken

#https://www.kaggle.com/c/tgs-salt-identification-challenge/discussion/67693
class ConvBn2d(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size=(3,3), stride=(1,1), padding=(1,1)):
        super(ConvBn2d, self).__init__()

        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, stride=stride, padding=padding, bias=False)
        self.bn = nn.BatchNorm2d(out_channels)

    def forward(self, z):
        x = self.conv(z)

level14taken

#https://www.kaggle.com/c/tgs-salt-identification-challenge/discussion/67693
def iou_metric(outputs, labels,logits=True):
    outputs,labels= size_correct(outputs,labels)
    outputs= (outputs>0).detach().cpu().numpy() if logits else (outputs>.5).detach().cpu().numpy() 
    labels= labels.detach().cpu().numpy()
    batch_size = outputs.shape[0]
    metric = 0.0
    for batch in range(batch_size):
        t, p = labels[batch], outputs[batch]
        true = np.sum(t)

level14taken

kf = StratifiedKFold(10,shuffle=True,random_state=seed)
for fold, (trn_idx,val_idx) in enumerate(kf.split(file_list,coverage)):
    print('****************************************************')
    print('*******************   fold  %d   *******************' % fold)
    print('****************************************************')

    file_list_train= [x for i,x in enumerate(file_list) if i in trn_idx]
    file_list_val= [x for i,x in enumerate(file_list) if i in val_idx]
    train = TGSSaltDataset(train_path, file_list_train,augment=transform_train)
    val = TGSSaltDataset(train_path, file_list_val,augment= transform_test)

level14taken

class ConvBn2d(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size=(3,3), stride=(1,1), padding=(1,1)):
        super(ConvBn2d, self).__init__()

        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, stride=stride, padding=padding, bias=False)
        self.bn = nn.BatchNorm2d(out_channels)

    def forward(self, z):
        x = self.conv(z)
        x = self.bn(x)

level14taken

def get_losslv_simple(model,inp):
  no_mask,final=model(inp[0])
  final=final*inp[-1].view(-1,1,1,1)
  mfl=nn.BCEWithLogitsLoss()
  lv=LovaszLoss()
  weights=[.1/2.1,2/2.1]
  loss=[mfl(no_mask.view(-1,1,1,1),inp[-1].view(-1,1,1,1)),lv(final,inp[1])]
  accuracy=acc(no_mask.reshape(-1),inp[-1],center=True)/len(inp[-1])
  loss=sum(x_i*y_i for x_i, y_i in zip(weights, loss))  
  return loss,accuracy,final

level14taken

def test_loop_fn(loader):
  loss_,iou_,acc_=0.,0.,0.,
  model.eval()
  with torch.no_grad():
    for image, mask in  loader:
      image = image.to(device)
      mask=mask.to(device)
      y_pred= model(image)
      loss = loss_fn(y_pred, mask)
      loss_+= loss.item()

level14taken

def train_loop_fn(loader):
  model.train()
  loss_,iou_,acc_=0.,0.,0.
  for x ,(image, mask) in enumerate(loader): 
    mask=mask.to(device)
    image= image.to(device)
    y_pred = model(image)
    loss = loss_fn(y_pred, mask)
    loss_+= loss.item()
    optimizer.zero_grad()

level14taken

test_transforms=PadIfNeeded(128,128,cv2.BORDER_REPLICATE), 
 Normalize(mean=(0,0,0),std=(1,1,1,))])
transform_train = Compose([
    HorizontalFlip(p=.5),
     Compose([RandomCrop(90,90),
     Resize(101,101)],p=.3),
    OneOf([RandomBrightness(.1),
    RandomContrast(.1),RandomGamma()],p=.2), 
    PadIfNeeded(256//2,256//2,cv2.BORDER_REPLICATE),  
    Normalize(mean=(0,0,0),std=(1,1,1,))])

level14taken

class TGSSaltDataset(data.Dataset):
    def __init__(self, root_path, file_list, is_test = False,augment= def_transforms,dpsv=False):
        self.is_test = is_test
        self.root_path = root_path
        self.file_list = file_list
        self.augment= augment
        self.dpsv= dpsv
    def __len__(self):
        return len(self.file_list)