Table of Contents
Shortcuts
generated/torch_uncertainty.baselines.classification.ResNetBaseline

ResNetBaseline

class torch_uncertainty.baselines.classification.ResNetBaseline(num_classes, in_channels, loss, version, arch, style='imagenet', num_estimators=1, dropout_rate=0.0, mixtype='erm', mixmode='elem', dist_sim='emb', kernel_tau_max=1.0, kernel_tau_std=0.5, mixup_alpha=0, cutmix_alpha=0, last_layer_dropout=False, groups=1, scale=None, alpha=None, gamma=1, rho=1.0, batch_repeat=1, ood_criterion='msp', log_plots=False, save_in_csv=False, calibration_set=None, eval_ood=False, eval_grouping_loss=False, pretrained=False)[source]

ResNet backbone baseline for classification providing support for various versions and architectures.

Parameters:

  • num_classes (int) – Number of classes to predict.

  • in_channels (int) – Number of input channels.

  • loss (nn.Module) – Training loss.

  • optim_recipe (Any) – optimization recipe, corresponds to what expect the LightningModule.configure_optimizers() method.

  • version (str) –

    Determines which ResNet version to use:

    • "std": original ResNet

    • "packed": Packed-Ensembles ResNet

    • "batched": BatchEnsemble ResNet

    • "masked": Masksemble ResNet

    • "mimo": MIMO ResNet

    • "mc-dropout": Monte-Carlo Dropout ResNet

  • arch (int) –

    Determines which ResNet architecture to use:

    • 18: ResNet-18

    • 32: ResNet-32

    • 50: ResNet-50

    • 101: ResNet-101

    • 152: ResNet-152

  • style (str, optional) – Which ResNet style to use. Defaults to

  • imagenet. –

  • num_estimators (int, optional) – Number of estimators in the ensemble. Only used if version is either "packed", "batched", "masked" or "mc-dropout" Defaults to None.

  • dropout_rate (float, optional) – Dropout rate. Defaults to 0.0.

  • mixtype (str, optional) – Mixup type. Defaults to "erm".

  • mixmode (str, optional) – Mixup mode. Defaults to "elem".

  • dist_sim (str, optional) – Distance similarity. Defaults to "emb".

  • kernel_tau_max (float, optional) – Maximum value for the kernel tau. Defaults to 1.0.

  • kernel_tau_std (float, optional) – Standard deviation for the kernel tau. Defaults to 0.5.

  • mixup_alpha (float, optional) – Alpha parameter for Mixup. Defaults to 0.

  • cutmix_alpha (float, optional) – Alpha parameter for CutMix. Defaults to 0.

  • groups (int, optional) – Number of groups in convolutions. Defaults to 1.

  • scale (float, optional) – Expansion factor affecting the width of the estimators. Only used if version is "masked". Defaults to None.

  • last_layer_dropout (bool) – whether to apply dropout to the last layer only.

  • groups – Number of groups in convolutions. Defaults to 1.

  • scale – Expansion factor affecting the width of the estimators. Only used if version is "masked". Defaults to None.

  • alpha (float, optional) – Expansion factor affecting the width of the estimators. Only used if version is "packed". Defaults to None.

  • gamma (int, optional) – Number of groups within each estimator. Only used if version is "packed" and scales with groups. Defaults to 1.

  • rho (float, optional) – Probability that all estimators share the same input. Only used if version is "mimo". Defaults to 1.

  • batch_repeat (int, optional) – Number of times to repeat the batch. Only used if version is "mimo". Defaults to 1.

  • ood_criterion (str, optional) – OOD criterion. Defaults to "msp". MSP is the maximum softmax probability, logit is the maximum logit, entropy is the entropy of the mean prediction, mi is the mutual information of the ensemble and vr is the variation ratio of the ensemble.

  • log_plots (bool, optional) – Indicates whether to log the plots or not. Defaults to False.

  • save_in_csv (bool, optional) – Indicates whether to save the results in a csv file or not. Defaults to False.

  • calibration_set (Callable, optional) – Calibration set. Defaults to None.

  • eval_ood (bool, optional) – Indicates whether to evaluate the OOD detection or not. Defaults to False.

  • eval_grouping_loss (bool, optional) – Indicates whether to evaluate the grouping loss or not. Defaults to False.

  • pretrained (bool, optional) – Indicates whether to use the pretrained weights or not. Only used if version is "packed". Defaults to False.

Raises:

ValueError – If version is not either "std", "packed", "batched", "masked" or "mc-dropout".

Returns:

ResNet baseline ready for training and evaluation.

Return type:

LightningModule