SEQ CARS196

Classes

class datasets.seq_cars196.MyCars196(root, train=True, transform=None, target_transform=None)

Bases: Dataset

N_CLASSES = 196

Overrides the CIFAR100 dataset to change the getitem function.

PREPROCESSING_TRANSFORM = Compose(     Resize(size=224, interpolation=bicubic, max_size=None, antialias=True)     CenterCrop(size=(224, 224)) )

static get_class_names()

load_and_preprocess_dataset(root, train_str='train')

class datasets.seq_cars196.SequentialCars196(args)

Bases: ContinualDataset

Sequential CARS196 Dataset. The images are loaded from deeplake, resized to 224x224, and store locally.

MEAN = (0.0, 0.0, 0.0)

NAME: str = 'seq-cars196'

N_CLASSES: int = 196

N_CLASSES_PER_TASK: int = [20, 20, 20, 20, 20, 20, 20, 20, 20, 16]

N_TASKS: int = 10

SETTING: str = 'class-il'

SIZE: Tuple[int] = (224, 224)

STD = (1.0, 1.0, 1.0)

TEST_TRANSFORM = Compose(     ToTensor()     Normalize(mean=(0.0, 0.0, 0.0), std=(1.0, 1.0, 1.0)) )

TRANSFORM = Compose(     RandomHorizontalFlip(p=0.5)     ToTensor()     Normalize(mean=(0.0, 0.0, 0.0), std=(1.0, 1.0, 1.0)) )

get_backbone()

get_batch_size()

get_class_names()

get_data_loaders()

Return type:

Tuple[DataLoader, DataLoader]

static get_denormalization_transform()

get_epochs()

static get_loss()

static get_normalization_transform()

static get_prompt_templates()

static get_transform()

Functions

datasets.seq_cars196.load_and_preprocess_cars196(train_str='train', names_only=False)

Loads data from deeplake and preprocesses it to be stored locally.

Parameters:

• train_str (str) – ‘train’ or ‘test’.

• names_only (bool) – If True, returns the class names only.

Returns:

If names_only is False, returns a tuple of data, targets, and class_idx_to_name

Return type:

Tuple[torch.Tensor, torch.Tensor, dict] | dict