If model is None: use pretrained models. Otherwise, use the provided models.

Calculate accuracy according to the prediction and target. Args: pred (torch.Tensor): The model prediction, shape (N, num_class)

1x1 convolution

3x3 convolution with padding

KNN estimation

Input: data the extracted embedding / features from the given dataset; {array-like, sparse matrix} of shape (n_samples, n_

Apply element-wise weight and reduce loss. Args: loss (Tensor): Element-wise loss. weight (Tensor): Element-wise weights.

merge dict ``a`` into dict ``b`` (non-inplace). Values in ``a`` will overwrite ``b``. ``b`` is copied first to avoid in-place modific

Expand onehot labels to match the size of prediction.

Parse iterable values in the string. All elements inside '()' or '[]' are treated as iterable values. Args: val (str): V

Substitute base variable placehoders to string, so that parsing would work.

Substitute variable strings to their actual values.

Optimize over the noise transition matrix T and prior P

r""" Compute the first-, second-, and third-order of consensus matrices. Args: cfg: configuration T : noise transition matrix

Count the consensus

Count the distribution of KNN Args: cfg: configuration dataset: the data for estimation sample: the index of samples

HOC estimator

Compute the loss of estimated consensus matrices

Randomly generate the label noise transition matrix (T) with given diagonal elements (diag). See ref: https://proceedings.mlr.press/

Get the corruption score. Lower score indicates the sample is more likely to be corrupted. Args: knn_labels_cnt: KNN labels label

Import modules from the given list of strings. Args: imports (list | str | None): The given module names to be imported. allow_fa

Flip classes according to transition probability matrix T.

Synthesize class-dependent label noise according to a random T

This function servers to preporcess your dataset for docta use. The code below will do some basic cleaning stuff, then return you the

Reduce loss as specified. Args: loss (Tensor): Elementwise loss tensor. reduction (str): Options are "none", "mean" and "sum".

Input: embed --> the extracted embedding / features from the given dataset; {array-like, sparse matrix} of shape (n_samples, n_featur

Construct the set of data that are likely to be corrupted.

Args: index (int): Index Returns: tuple: (img, (target, noisy_label), index). target: clean lab