Training ImageNet-1K

[nlrahimi]

Hi, Thanks for the amazing work. I enjoyed reading your paper. Running experiments I wanted to train ImageNet-1k using some baseline methods you have kindly shared the implementation for. However, I noticed ImageNet-100 is loaded in a pickle file, unlike frameworks that use dataloader for imagenet to load batches in a few phases. The problem is that when it comes to ImageNet1k its not possible to load all data at once. My question is how can I use dataloaders in Mini_ImageNet class in continuum/dataset_scripts structure, in a way that it does not disrupt other functions in relevant codes?

` class Mini_ImageNet(DatasetBase):

 def __init__(self, scenario, params): 

 def download_load(self):

 def new_task(self, cur_task, **kwargs):

        elif self.scenario == 'nc':

            labels = self.task_labels[cur_task]

            x_train, y_train = load_task_with_labels(self.train_data, self.train_label, labels)

        return x_train, y_train, labels

`