The tensor of images needs to be a fixed size, (N, C, H, W), for an image in the tensor with size (H', W') and another with (H, W)

Question

So i am trying to train a model in a specific dataset, Camus Dataset, but i am new to pytorch so i am a little it confused. I used the Camus Dataset code of https://www.kaggle.com/code/sontungtran/camus-eda . But when i run a part of code for display image i see that some images have different sizes of others, so when i go down the step for train the model and with batch size, i always have a problem.

So i used this code

class CamusIterator(Dataset):
def __init__(
self,
data_type,
global_transforms = [],
augment_transforms =[],
data_file = []

    ):


    super(CamusIterator, self).__init__()
    
    train_file='/content/drive/MyDrive/Echo_Segmentation/tries/training_set'
    test_file='/content/drive/MyDrive/Echo_Segmentation/tries/testing_set'
    
    if data_type == 'train':
        data_file = train_file
        self.images = os.listdir(train_file)
    elif data_type == 'test':
        data_file = test_file
        self.images = os.listdir(test_file)
    else:
        raise Exception('Wrong data_type for CamusIterator')

    self.data_type = data_type
    self.data_file = data_file
    self.global_transforms = global_transforms
    self.augment_transforms = augment_transforms

def __read_image( self, patient_file, suffix ):
    image_file = '{}/{}/{}'.format( self.data_file, patient_file, patient_file+suffix )
    # # Stolen from a StackOverflow answer
    # # https://stackoverflow.com/questions/37290631/reading-png-raw-format-in-python
    image = sitk.GetArrayFromImage( sitk.ReadImage(image_file, sitk.sitkFloat32) )
    #size = [1232,748]
    # plt.figure(figsize=(20,16))
    # plt.gray()
    # plt.subplots_adjust(0,0,1,1,0.01,0.01)
    # for i in range(image.shape[0]):
    #     plt.subplot(5,6,i+1), plt.imshow(image[i]), plt.axis('off')
    #     # use plt.savefig(...) here if you want to save the images as .jpg, e.g.
    #     plt.savefig('image.jpg')
    # #plt.show()    
    print(image.shape)
    return image
    
def __read_info( self, data_file ):
    info = {}
    with open( data_file, 'r' ) as f:
        for line in f.readlines():
            info_type, info_details = line.strip( '\n' ).split( ': ' )
            info[ info_type ] = info_details
    return info    

def __len__( self ):
    return len( os.listdir(self.data_file) )

def __getitem__( self, index ):
    #patient_file = 'patient{}'.format( f'{index+1:04}' ) # patient{0001}, patient{0002}, etc
    #patient_file= os.listdir(self.data_file)[index]
    patient_file= os.listdir(self.data_file)[index]

    image_2CH_ED = self.__read_image( patient_file, '_2CH_ED.mhd' )
    image_2CH_ES = self.__read_image( patient_file, '_2CH_ES.mhd' )
    image_4CH_ED = self.__read_image( patient_file, '_4CH_ED.mhd' )
    image_4CH_ES = self.__read_image( patient_file, '_4CH_ES.mhd' )
    image_2CH_sequence = self.__read_image( patient_file, '_2CH_sequence.mhd' )
    image_4CH_sequence = self.__read_image( patient_file, '_4CH_sequence.mhd' )

    
    if self.data_type == 'train':
    #for train in self.data_type:
        image_2CH_ED_gt = self.__read_image( patient_file, '_2CH_ED_gt.mhd' )
        image_2CH_ES_gt = self.__read_image( patient_file, '_2CH_ES_gt.mhd' )
        image_4CH_ED_gt = self.__read_image( patient_file, '_4CH_ED_gt.mhd' )
        image_4CH_ES_gt = self.__read_image( patient_file, '_4CH_ES_gt.mhd' )

    info_2CH = self.__read_info( '{}/{}/{}'.format(self.data_file, patient_file, 'Info_2CH.cfg') )
    info_4CH = self.__read_info( '{}/{}/{}'.format(self.data_file, patient_file, 'Info_4CH.cfg') )    

    
    if self.data_type == 'train':
        data = {
            'patient': patient_file,
            '2CH_ED': image_2CH_ED,
            '2CH_ES': image_2CH_ES,
            '4CH_ED': image_4CH_ED,
            '4CH_ES': image_4CH_ES,
            '2CH_sequence': image_2CH_sequence,
            '4CH_sequence': image_4CH_sequence,
            '2CH_ED_gt': image_2CH_ED_gt,
            '2CH_ES_gt': image_2CH_ES_gt,
            '4CH_ED_gt': image_4CH_ED_gt,
            '4CH_ES_gt': image_4CH_ES_gt,
            'info_2CH': info_2CH,    # Dictionary of infos
            'info_4CH': info_4CH     # Dictionary of infos 
        }
    elif self.data_type == 'test':
        data = {
            'patient': patient_file,
            '2CH_ED': image_2CH_ED,
            '2CH_ES': image_2CH_ES,
            '4CH_ED': image_4CH_ED,
            '4CH_ES': image_4CH_ES,
            '2CH_sequence': image_2CH_sequence,
            '4CH_sequence': image_4CH_sequence,
            'info_2CH': info_2CH,   # Dictionary of infos
            'info_4CH': info_4CH   # Dictionary of infos
        }
    # Transforms
    for transform in self.global_transforms:
        data = transform(data,self.data_type)
    for transform in self.augment_transforms:
        data = transform(data)    
        
    return data

def __iter__( self ):
    for i in range( len(self) ):
        yield self[ i ]

and then after this code i used:

def display_image(patient,fields):
 #fields=['2CH_ED', '2CH_ES', '4CH_ED', '4CH_ES','2CH_ED_gt', '2CH_ES_gt', '4CH_ED_gt', '4CH_ES_gt']
   fig = plt.figure(figsize=(40,60))
   for i,fname in enumerate(fields):
       test = fig.add_subplot(1, len(fields), i + 1)
       image = patient[fname]
       # print(len(np.unique(image)))
       plt.axis('off')
       # print(image.shape)
       plt.imshow(image.squeeze(),cmap = 'gray')
       test.set_title(fname)
   fig.tight_layout()
   plt.show()

for i in train_file:
     fields=[#'2CH_ED',
              '2CH_ED_gt',# '2CH_ES',
              '2CH_ES_gt',# '4CH_ED',
              '4CH_ED_gt',# '4CH_ES',
              '4CH_ES_gt']
     display_image(i,fields)
     break

and i take as a result something like that (i will not put all the values i get cause a lot of them are the same):

(1, 779, 417)
(1, 908, 551)
(1, 779, 417)
(1, 908, 582)
(1, 843, 512)
(1, 908, 582)
(1, 1168, 708)

so the images of my dataset have different (H,W). How can i change that? i gave try some things as you can see from the comment lines(this is just an example) but i can't figure out the right solution. Should i put the images and the masks in seperate ways somehow, and if yes how do i do that?