Linear rampup. Args: current (int): current step. rampup_length (int): maximum step.

Loads pretrained weights, and downloads if loading for the first time.

r"""Load pretrianed weights to model. Features:: - Incompatible layers (unmatched in name or size) will be ignored. - Can automat

r"""Copies parameters and buffers from :attr:`state_dict` into this module and its descendants. If :attr:`strict` is ``True``, then the keys o

r"""Open specified layers in model for training while keeping other layers frozen. Args: model (nn.Module): neural net model.

Read image from path using ``PIL.Image``. Args: path (str): path to an image. Returns: PIL image

r"""Resume training from a checkpoint. This will load (1) model weights and (2) ``state_dict`` of optimizer if ``optimizer`` is not None.

Round number of filters based on depth multiplier.

r"""Save checkpoint. Args: state (dict): dictionary. save_dir (str): directory to save checkpoint. is_best (bool, optiona

Sets swish function as memory efficient (for training) or standard (for export)

Sharpening probability with a temperature. Args: p (torch.Tensor): probability matrix (batch_size, n_classes) temperature (float)

Shuffle order.

Exponential Rampup https://arxiv.org/abs/1610.02242

Split a dataset, i.e. a list of Datum objects, into domain-specific groups stored in a dictionary. Args: data_source (lis

Split a dataset, i.e. a list of Datum objects, into class-specific groups stored in a dictionary. Args: data_source (list

Spatial transformer network.

Convert to a list.

Modified cost for logarithmic updates. Eq: M_{ij} = (-c_{ij} + u_i + v_j) / epsilon

Args: img (Tensor): tensor image of size (C, H, W). Returns: Tensor: image with n_holes of dimension