Skip to content

imagenet

Bases: vision_dataloader

A dataloader for the ImageNet dataset.

Handles loading and preprocessing of the ImageNet dataset.

Attributes:

Name Type Description
name str

Name of the dataloader, default is 'imagenet'.
train_batch_size int

Batch size for training data.
test_batch_size int

Batch size for testing data.

Methods:

Name Description
__init__

Initializes the ImageNet dataloader.
load

Loads and preprocesses the ImageNet dataset.
Source code in tinybig/data/vision_dataloader.py 
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
class imagenet(vision_dataloader):
    """
    A dataloader for the ImageNet dataset.

    Handles loading and preprocessing of the ImageNet dataset.

    Attributes
    ----------
    name : str
        Name of the dataloader, default is 'imagenet'.
    train_batch_size : int
        Batch size for training data.
    test_batch_size : int
        Batch size for testing data.

    Methods
    -------
    __init__(...)
        Initializes the ImageNet dataloader.
    load(...)
        Loads and preprocesses the ImageNet dataset.
    """
    def __init__(self, name='imagenet', train_batch_size: int = 64, test_batch_size: int = 64):
        """
        Initializes the ImageNet dataloader.

        Parameters
        ----------
        name : str, optional
            Name of the dataloader, default is 'imagenet'.
        train_batch_size : int, optional
            Batch size for training data, default is 64.
        test_batch_size : int, optional
            Batch size for testing data, default is 64.

        Returns
        -------
        None
        """
        super().__init__(name=name, train_batch_size=train_batch_size, test_batch_size=test_batch_size)

    # @staticmethod
    # def flatten(x):
    #     x = torch.flatten(x)
    #     return x.view(-1)

    def load(self, cache_dir='./data/', with_transformation: bool = True, *args, **kwargs):
        """
        Loads and preprocesses the ImageNet dataset.

        Parameters
        ----------
        cache_dir : str, optional
            Directory to cache the dataset, default is './data/'.
        with_transformation : bool, optional
            Whether to load the training or testing dataset with transformation, default is True.
        *args, **kwargs
            Additional arguments for dataset loading.

        Returns
        -------
        dict
            A dictionary containing the train and test dataloaders.
        """
        if with_transformation:
            imagenet_transform = transforms.Compose([
                transforms.Resize(256),
                transforms.RandomResizedCrop(224),
                #transforms.RandomHorizontalFlip(),
                transforms.ToTensor(),
                transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
                torch.flatten
            ])
        else:
            imagenet_transform = Compose([
                transforms.ToTensor(),
            ])

        train_loader = DataLoader(
            ImageNet(root=cache_dir, split='train', transform=imagenet_transform),
            batch_size=self.train_batch_size, shuffle=True)

        test_loader = DataLoader(
            ImageNet(root=cache_dir, split='val', transform=imagenet_transform),
            batch_size=self.test_batch_size, shuffle=False)

        return {'train_loader': train_loader, 'test_loader': test_loader}

__init__(name='imagenet', train_batch_size=64, test_batch_size=64)

Initializes the ImageNet dataloader.

Parameters:

Name Type Description Default
name str

Name of the dataloader, default is 'imagenet'.

 'imagenet'
train_batch_size int

Batch size for training data, default is 64.

 64
test_batch_size int

Batch size for testing data, default is 64.

 64

Returns:

Type Description
None
Source code in tinybig/data/vision_dataloader.py 
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
def __init__(self, name='imagenet', train_batch_size: int = 64, test_batch_size: int = 64):
    """
    Initializes the ImageNet dataloader.

    Parameters
    ----------
    name : str, optional
        Name of the dataloader, default is 'imagenet'.
    train_batch_size : int, optional
        Batch size for training data, default is 64.
    test_batch_size : int, optional
        Batch size for testing data, default is 64.

    Returns
    -------
    None
    """
    super().__init__(name=name, train_batch_size=train_batch_size, test_batch_size=test_batch_size)

load(cache_dir='./data/', with_transformation=True, *args, **kwargs)

Loads and preprocesses the ImageNet dataset.

Parameters:

Name Type Description Default
cache_dir str

Directory to cache the dataset, default is './data/'.

 './data/'
with_transformation bool

Whether to load the training or testing dataset with transformation, default is True.

 True
*args

Additional arguments for dataset loading.

 ()
**kwargs

Additional arguments for dataset loading.

 ()

Returns:

Type Description
dict

A dictionary containing the train and test dataloaders.
Source code in tinybig/data/vision_dataloader.py 
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
def load(self, cache_dir='./data/', with_transformation: bool = True, *args, **kwargs):
    """
    Loads and preprocesses the ImageNet dataset.

    Parameters
    ----------
    cache_dir : str, optional
        Directory to cache the dataset, default is './data/'.
    with_transformation : bool, optional
        Whether to load the training or testing dataset with transformation, default is True.
    *args, **kwargs
        Additional arguments for dataset loading.

    Returns
    -------
    dict
        A dictionary containing the train and test dataloaders.
    """
    if with_transformation:
        imagenet_transform = transforms.Compose([
            transforms.Resize(256),
            transforms.RandomResizedCrop(224),
            #transforms.RandomHorizontalFlip(),
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
            torch.flatten
        ])
    else:
        imagenet_transform = Compose([
            transforms.ToTensor(),
        ])

    train_loader = DataLoader(
        ImageNet(root=cache_dir, split='train', transform=imagenet_transform),
        batch_size=self.train_batch_size, shuffle=True)

    test_loader = DataLoader(
        ImageNet(root=cache_dir, split='val', transform=imagenet_transform),
        batch_size=self.test_batch_size, shuffle=False)

    return {'train_loader': train_loader, 'test_loader': test_loader}