Initializes our module using our above struct

Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text` and `kwargs` arguments

Args: tensor (Tensor): Tensor image of size (C, H, W) to be normalized. Returns: Tensor: Normalized image.

Args: sample_info(dict): Returned dict from antmmf.datasets.database.features_database.FeaturesDatabase.

Collate examples for supervised fine-tuning.

Args: sample_info(dict): Returned dict from antmmf.datasets.database.features_database.FeaturesDatabase.

Args: sample_info(dict): Returned dict from antmmf.datasets.database.features_database.FeaturesDatabase.

Args: sample_info(dict): Returned dict from antmmf.datasets.database.features_database.FeaturesDatabase.

:param tensor: bsz, c, h, w / c, h, w :param kwargs: :return:

Args: img (torch.tensor): Image to be scaled. Returns: torch.tensor: Rescaled image.

Method to be called everytime you need to check whether to early stop or not Arguments: iteration {number}: Curre

Call requires item to have either "image" attribute or either "feature" attribute. If "image" is present, it will processed using an i

Main function of the processor. Takes in a dict and returns back a dict Args: item (Dict, List): Some item that needs to

Args: item: must have 'hier_label' key, whose corresponding value is a label or hier-label with '-' concatenate

Call requires item to have either "image" attribute or either "feature" attribute. If "image" is present, it will processed using an i

Args: data: {'image': image_arr, 'labels': annotations, # Nx5}, image_arr is BGR or RGB for

Args: data: {'image': image_arr, # pil RGB format, 'target': None or {'bbox': annotations, # [x1,y1,x2,y2], Nx

Takes in dict with answers or answers_tokens, and returns back a dict with answers (processed), "answers_indices" which point to indic