DD-Ranking API Documentation

AugmentationRobustScore

A class for evaluating the performance of a dataset distillation method with soft labels. User is able to modify the attributes as needed.

Parameters

• config(Optional[Config]): Config object for specifying all attributes. See config for more details.
• dataset(str): Name of the real dataset.
• real_data_path(str): Path to the real dataset.
• ipc(int): Images per class.
• model_name(str): Name of the surrogate model. See models for more details.
• label_type(str): Type of label representation. soft for soft labels, hard for hard labels.
• soft_label_mode(str): Number of soft labels per image. S for single soft label, M for multiple soft labels.
• soft_label_criterion(str): Loss function for using soft labels. Currently supports kl for KL divergence, sce for soft cross-entropy, and mse_gt for MSEGT loss introduced in EDC.
• loss_fn_kwargs(dict): Keyword arguments for the loss function, e.g. temperature and scale_loss for KL and SCE loss, and mse_weight and ce_weight for MSEGT loss.
• data_aug_func(str): Data augmentation function used during training. Currently supports dsa, cutmix, mixup. See augmentations for more details.
• aug_params(dict): Parameters for the data augmentation function.
• optimizer(str): Name of the optimizer. Currently supports torch-based optimizers - sgd, adam, and adamw.
• lr_scheduler(str): Name of the learning rate scheduler. Currently supports torch-based schedulers - step, cosine, lambda_step, and cosineannealing.
• weight_decay(float): Weight decay for the optimizer.
• momentum(float): Momentum for the optimizer.
• step_size(int): Step size for the learning rate scheduler.
• use_zca(bool): Whether to use ZCA whitening.
• num_eval(int): Number of evaluations to perform.
• im_size(tuple): Size of the images.
• num_epochs(int): Number of epochs to train.
• batch_size(int): Batch size for the model training.
• stu_use_torchvision(bool): Whether to use torchvision to initialize the student model.
• tea_use_torchvision(bool): Whether to use torchvision to initialize the teacher model.
• teacher_dir(str): Path to the teacher model.
• teacher_model_names(list): List of teacher model names.
• random_data_format(str): Format of the random data, tensor or image.
• random_data_path(str): Path to save the random data.
• num_workers(int): Number of workers for data loading.
• save_path(Optional[str]): Path to save the results.
• custom_train_trans(Optional[Callable]): Custom transformation function when loading synthetic data. Only support torchvision transformations. See torchvision-based transformations for more details.
• custom_val_trans(Optional[Callable]): Custom transformation function when loading test dataset. Only support torchvision transformations. See torchvision-based transformations for more details.
• device(str): Device to use for evaluation, cuda or cpu.
• dist(bool): Whether to use distributed training.

Methods

Examples

with config file:

>>> config = Config('/path/to/config.yaml')
>>> evaluator = AugmentationRobustScore(config=config)
# load image and soft labels
>>> image_tensor, soft_labels = ... 
# compute metrics
>>> evaluator.compute_metrics(image_tensor=image_tensor, soft_labels=soft_labels)
# alternatively, provide image path
>>> evaluator.compute_metrics(image_path='path/to/image/folder/', soft_labels=soft_labels) 

with keyword arguments:

>>> evaluator = AugmentationRobustScore(
...     dataset='ImageNet1K',
...     real_data_path='./dataset/',
...     ipc=10,
...     model_name='ResNet-18-BN',
...     label_type='soft',
...     soft_label_mode='M',
...     soft_label_criterion='kl',
...     loss_fn_kwargs={
...         "temperature": 30.0,
...         "scale_loss": False,
...     },
...     data_aug_func='mixup',
...     aug_params={
...         "mixup_p": 0.8,
...     },
...     optimizer='adamw',
...     lr_scheduler='cosine',
...     num_epochs=300,
...     weight_decay=0.0005,
...     momentum=0.9,
...     use_zca=False,
...     stu_use_torchvision=True,
...     tea_use_torchvision=True,
...     num_workers=4,
...     save_path='./results',
...     random_data_format='image',
...     random_data_path='./random_data',
...     teacher_dir='./teacher_models',
...     teacher_model_names=['ResNet-18-BN'],
...     num_eval=5,
...     device='cuda'
... )
# load image and soft labels
>>> image_tensor, soft_labels = ... 
# compute metrics
>>> evaluator.compute_metrics(image_tensor=image_tensor, soft_labels=soft_labels)
# alternatively, provide image path
>>> evaluator.compute_metrics(image_path='path/to/image/folder/', soft_labels=soft_labels)