.rst .pdf

mlreco.models.layers.cluster_cnn.losses.misc module

mlreco.models.layers.cluster_cnn.losses.misc.logit_fn(input, eps=1e-06)

mlreco.models.layers.cluster_cnn.losses.misc.unique_label_torch(label)

mlreco.models.layers.cluster_cnn.losses.misc.iou_batch(pred: torch.BoolTensor, labels: torch.BoolTensor, eps=0.0)

pred: N x C labels: N x C (one-hot)

class mlreco.models.layers.cluster_cnn.losses.misc.VectorEstimationLoss(cfg, name='vector_estimation_loss')

Bases: torch.nn.modules.module.Module

__init__(cfg, name='vector_estimation_loss')

Initializes internal Module state, shared by both nn.Module and ScriptModule.

compute_loss_single_graph(vec_pred, pos)

INPUTS:

• x (N x 4) : sin/cos predictions for phi and theta.

• pos (N x 3 Tensor): spatial coordinates (batch, semantic_id)

forward(graph)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

__module__ = 'mlreco.models.layers.cluster_cnn.losses.misc'

training: bool

class mlreco.models.layers.cluster_cnn.losses.misc.BinaryLogDiceLoss(gamma=1)

Bases: torch.nn.modules.module.Module

__init__(gamma=1)

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(logits, targets, eps=1e-06)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

__module__ = 'mlreco.models.layers.cluster_cnn.losses.misc'

training: bool

class mlreco.models.layers.cluster_cnn.losses.misc.IoUScore

Bases: torch.nn.modules.module.Module

__init__()

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(y_pred, y_true)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

__module__ = 'mlreco.models.layers.cluster_cnn.losses.misc'

training: bool

class mlreco.models.layers.cluster_cnn.losses.misc.BinaryCELogDiceLoss(gamma=0.3, w_ce=0.2, w_dice=0.8)

Bases: torch.nn.modules.module.Module

__init__(gamma=0.3, w_ce=0.2, w_dice=0.8)

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(logits, targets, weight=None, eps=0.001, reduction='none')

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

__module__ = 'mlreco.models.layers.cluster_cnn.losses.misc'

training: bool

class mlreco.models.layers.cluster_cnn.losses.misc.MincutLoss(mincut_weight=1.0, **kwargs)

Bases: mlreco.models.layers.cluster_cnn.losses.misc.BinaryCELogDiceLoss

__init__(mincut_weight=1.0, **kwargs)

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(logits, targets, weight=None, eps=0.001, reduction='none')

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

__module__ = 'mlreco.models.layers.cluster_cnn.losses.misc'

training: bool

class mlreco.models.layers.cluster_cnn.losses.misc.LovaszHingeLoss(reduction='none')

Bases: torch.nn.modules.loss._Loss

__init__(reduction='none')

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(logits, targets)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

__module__ = 'mlreco.models.layers.cluster_cnn.losses.misc'

reduction: str

class mlreco.models.layers.cluster_cnn.losses.misc.LovaszSoftmaxWithLogitsLoss(reduction='none')

Bases: torch.nn.modules.loss._Loss

__init__(reduction='none')

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(logits, targets)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

__module__ = 'mlreco.models.layers.cluster_cnn.losses.misc'

reduction: str

mlreco.models.layers.cluster_cnn.losses.misc.find_cluster_means(features, labels)

For a given image, compute the centroids mu_c for each cluster label in the embedding space. Inputs:

features (torch.Tensor): the pixel embeddings, shape=(N, d) where N is the number of pixels and d is the embedding space dimension. labels (torch.Tensor): ground-truth group labels, shape=(N, )

Returns

(n_c, d) tensor where n_c is the number of distinct instances. Each row is a (1,d) vector corresponding to the coordinates of the i-th centroid.

Return type

cluster_means (torch.Tensor)

mlreco.models.layers.cluster_cnn.losses.misc.intra_cluster_loss(features, cluster_means, labels, margin=1.0)

mlreco.models.layers.cluster_cnn.losses.misc.inter_cluster_loss(cluster_means, margin=0.2)

mlreco.models.layers.cluster_cnn.losses.misc.regularization_loss(cluster_means)

mlreco.models.layers.cluster_cnn.losses.misc.margin_smoothing_loss(sigma, sigma_means, labels, margin=0)

mlreco.models.layers.cluster_cnn.losses.misc.get_probs(embeddings, margins, labels, eps=1e-06)

mlreco.models.layers.cluster_cnn.losses.misc.multivariate_kernel(centroid, log_sigma, Lprime, eps=1e-08)

mlreco.models.layers.cluster_cnn.losses.misc.bhattacharyya_distance_matrix(v1, v2, eps=1e-08)

mlreco.models.layers.cluster_cnn.losses.misc.squared_distances(v1, v2)

mlreco.models.layers.cluster_cnn.losses.misc.bhattacharyya_coeff_matrix(v1, v2, eps=1e-06)

mlreco.models.layers.cluster_cnn.losses.misc.get_graphspice_logits(sp_emb, ft_emb, cov, groups, sp_centroids, ft_centroids, eps=0.001, compute_accuracy=True)

class mlreco.models.layers.cluster_cnn.losses.misc.FocalLoss(alpha=1, gamma=2, logits=False, reduce=True)

Bases: torch.nn.modules.module.Module

Original Paper: https://arxiv.org/abs/1708.02002 Implementation: https://www.kaggle.com/c/tgs-salt-identification-challenge/discussion/65938

__init__(alpha=1, gamma=2, logits=False, reduce=True)

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(inputs, targets)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

__module__ = 'mlreco.models.layers.cluster_cnn.losses.misc'

training: bool

class mlreco.models.layers.cluster_cnn.losses.misc.WeightedFocalLoss(alpha=1, gamma=2, logits=False, reduce=True)

Bases: mlreco.models.layers.cluster_cnn.losses.misc.FocalLoss

__init__(alpha=1, gamma=2, logits=False, reduce=True)

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(inputs, targets)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

__module__ = 'mlreco.models.layers.cluster_cnn.losses.misc'

training: bool

mlreco.models.layers.cluster_cnn.losses.lovasz module mlreco.models.layers.cluster_cnn.losses.radius_nnloss module