Update the confusion matrices :param pred: B x T x H x W :param gt: B x T x H x W :return: None

Lightweight Temporal Attention Encoder (L-TAE) for image time series. Attention-based sequence encoding that maps a sequence of image

Returns the ground truth label of the given split.

Initializes the Losses object. Args: mix (dict): dictionary with keys "loss_name" and values weight

Multi-temporal U-TAE architecture for multi-stamp spatio-temporal encoding of satellite image time series. Args: input_di

Returns the names of the model parameters that are not inside a forbidden layer. Taken from HuggingFace transformers.

Initializes the losses. Args: mix (dict): dictionary with keys "loss_name" and values weight

Needs to be implemented in subclass.

Returns an item from the dataset. Args: i (int): index of the item Returns: dict: dictionary with keys "data",

Collate function for the dataloader. Args: batch (list): list of dictionaries with keys "data", "gt", "positions" and "idx" Returns:

Args: x: dict that contains "logits": torch.Tensor BxTxKxHxW y: dict that contains "gt": torch.Tensor BxTxHxW

Computes the losses. Args: x: dict that contains "logits": torch.Tensor BxTxKxHxW y: dict that contains "gt": torch.Te

Setup the datamodule. Args: stage (str): stage of the datamodule Is be one of "fit" or "test" or None