create_dl_layer_loss_focalT_create_dl_layer_loss_focalCreateDlLayerLossFocalCreateDlLayerLossFocalcreate_dl_layer_loss_focal (Operator)

Name

create_dl_layer_loss_focalT_create_dl_layer_loss_focalCreateDlLayerLossFocalCreateDlLayerLossFocalcreate_dl_layer_loss_focal — Create a focal loss layer.

Signature

Description

The operator create_dl_layer_loss_focalcreate_dl_layer_loss_focalCreateDlLayerLossFocalCreateDlLayerLossFocalCreateDlLayerLossFocalcreate_dl_layer_loss_focal creates a focal loss layer whose handle is returned in DLLayerLossFocalDLLayerLossFocalDLLayerLossFocalDLLayerLossFocalDLLayerLossFocaldllayer_loss_focal. See the reference cited below for further information about its definition and parameter meanings.

This layer expects multiple layers as input:

• DLLayerInputDLLayerInputDLLayerInputDLLayerInputDLLayerInputdllayer_input: Specifies the prediction (e.g., a sigmoid or softmax layer).

• DLLayerTargetDLLayerTargetDLLayerTargetDLLayerTargetDLLayerTargetdllayer_target: Specifies the target sequences (originating from the ground truth information).

• DLLayerWeightsDLLayerWeightsDLLayerWeightsDLLayerWeightsDLLayerWeightsdllayer_weights: Specifies the weight sequences. This parameter is optional. If an empty tuple [] is passed for all values the weighting factor 1.0 is used.

• DLLayerNormalizationDLLayerNormalizationDLLayerNormalizationDLLayerNormalizationDLLayerNormalizationdllayer_normalization: Specifies the factor to normalize the loss. This parameter is optional, it can be given by the layer handle as value or ignored handing over an empty tuple [].

The parameter LayerNameLayerNameLayerNameLayerNamelayerNamelayer_name sets an individual layer name. Note that if creating a model using create_dl_modelcreate_dl_modelCreateDlModelCreateDlModelCreateDlModelcreate_dl_model each layer of the created network must have a unique name.

The parameter LossWeightLossWeightLossWeightLossWeightlossWeightloss_weight is a overall loss weight if there are multiple losses in the network.

The parameter GammaGammaGammaGammagammagamma is the exponent of the focal factor.

The parameter ClassWeightsClassWeightsClassWeightsClassWeightsclassWeightsclass_weights defines class specific weights. All loss contributions of foreground samples of a class are weighted with the given factor. The background samples are weighted by 1 - ClassWeightsClassWeightsClassWeightsClassWeightsclassWeightsclass_weights. Typically, this is set to 1.0/(Number of samples of the class). Note, the length of this array has to be either 1, then its broadcasted to the number of classes, or it has to correspond to the number of classes. The default value [] corresponds to a factor of 0.5 for all classes. Note, if the number of classes are changed on a network then the number of class specific weights are also adapted and reset with the default value 0.5 for each class.

The parameter TypeTypeTypeTypetypetype sets the focal loss options:

'focal_binary'"focal_binary""focal_binary""focal_binary""focal_binary""focal_binary":

Focal loss.

'sigmoid_focal_binary'"sigmoid_focal_binary""sigmoid_focal_binary""sigmoid_focal_binary""sigmoid_focal_binary""sigmoid_focal_binary":

Focal loss fused with sigmoid.

The following generic parameters GenParamNameGenParamNameGenParamNameGenParamNamegenParamNamegen_param_name and the corresponding values GenParamValueGenParamValueGenParamValueGenParamValuegenParamValuegen_param_value are supported:

'is_inference_output'"is_inference_output""is_inference_output""is_inference_output""is_inference_output""is_inference_output":

Determines whether apply_dl_modelapply_dl_modelApplyDlModelApplyDlModelApplyDlModelapply_dl_model will include the output of this layer in the dictionary DLResultBatchDLResultBatchDLResultBatchDLResultBatchDLResultBatchdlresult_batch even without specifying this layer in OutputsOutputsOutputsOutputsoutputsoutputs ('true'"true""true""true""true""true") or not ('false'"false""false""false""false""false").

Default: 'false'"false""false""false""false""false"

Certain parameters of layers created using this operator create_dl_layer_loss_focalcreate_dl_layer_loss_focalCreateDlLayerLossFocalCreateDlLayerLossFocalCreateDlLayerLossFocalcreate_dl_layer_loss_focal can be set and retrieved using further operators. The following tables give an overview, which parameters can be set using set_dl_model_layer_paramset_dl_model_layer_paramSetDlModelLayerParamSetDlModelLayerParamSetDlModelLayerParamset_dl_model_layer_param and which ones can be retrieved using get_dl_model_layer_paramget_dl_model_layer_paramGetDlModelLayerParamGetDlModelLayerParamGetDlModelLayerParamget_dl_model_layer_param or get_dl_layer_paramget_dl_layer_paramGetDlLayerParamGetDlLayerParamGetDlLayerParamget_dl_layer_param. Note, the operators set_dl_model_layer_paramset_dl_model_layer_paramSetDlModelLayerParamSetDlModelLayerParamSetDlModelLayerParamset_dl_model_layer_param and get_dl_model_layer_paramget_dl_model_layer_paramGetDlModelLayerParamGetDlModelLayerParamGetDlModelLayerParamget_dl_model_layer_param require a model created by create_dl_modelcreate_dl_modelCreateDlModelCreateDlModelCreateDlModelcreate_dl_model.

Execution Information

• Multithreading type: reentrant (runs in parallel with non-exclusive operators).
• Multithreading scope: global (may be called from any thread).
• Processed without parallelization.

Parameters

References

T. Lin, P. Goyal, R. Girshick, K. He and P. Dollar, "Focal Loss for Dense Object Detection," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 42, no. 2, pp. 318-327, 1 Feb. 2020, doi: 10.1109/TPAMI.2018.2858826.

Module

Deep Learning Training