transform_dl_sample_batchT_transform_dl_sample_batchTransformDlSampleBatchTransformDlSampleBatchtransform_dl_sample_batch (Operator)

Name

transform_dl_sample_batchT_transform_dl_sample_batchTransformDlSampleBatchTransformDlSampleBatchtransform_dl_sample_batch — Apply a pipeline of transforms on a set of images.

Signature

transform_dl_sample_batch( : : TransformPipeline, DLSampleBatch : DLResultBatch)

Description

transform_dl_sample_batchtransform_dl_sample_batchTransformDlSampleBatchTransformDlSampleBatchtransform_dl_sample_batch transforms a batch of samples prior to their use in train_dl_model_batchtrain_dl_model_batchTrainDlModelBatchTrainDlModelBatchtrain_dl_model_batch and apply_dl_modelapply_dl_modelApplyDlModelApplyDlModelapply_dl_model. The transformations defined in TransformPipelineTransformPipelineTransformPipelinetransformPipelinetransform_pipeline may include adapting the image dimensions, scaling the gray value range, and applying data augmentation methods.

Besides the image data keys image and image_id contained in DLSampleBatchDLSampleBatchDLSampleBatchDLSampleBatchdlsample_batch, annotations are also transformed if present. Currently, only keys for object detection models (including instance segmentation) are supported. The behavior for 'instance_type'"instance_type""instance_type""instance_type""instance_type" = 'rectangle2'"rectangle2""rectangle2""rectangle2""rectangle2" can be influenced by the parameters 'ignore_direction'"ignore_direction""ignore_direction""ignore_direction""ignore_direction" and 'class_ids_no_orientation'"class_ids_no_orientation""class_ids_no_orientation""class_ids_no_orientation""class_ids_no_orientation", see get_dl_transform_pipeline_paramget_dl_transform_pipeline_paramGetDlTransformPipelineParamGetDlTransformPipelineParamget_dl_transform_pipeline_param. If DLSampleBatchDLSampleBatchDLSampleBatchDLSampleBatchdlsample_batch contains annotation keys for other model types, an error may be raised.

In addition to transformed keys, DLResultBatchDLResultBatchDLResultBatchDLResultBatchdlresult_batch contains

'transformed_full_domain'"transformed_full_domain""transformed_full_domain""transformed_full_domain""transformed_full_domain": Transformation of the rectangular full domain of the input image, clipped to output.
'back_transform'"back_transform""back_transform""back_transform""back_transform": Transformation that can be used together with projective_trans_pixelprojective_trans_pixelProjectiveTransPixelProjectiveTransPixelprojective_trans_pixel or projective_trans_regionprojective_trans_regionProjectiveTransRegionProjectiveTransRegionprojective_trans_region to transform inference results of apply_dl_modelapply_dl_modelApplyDlModelApplyDlModelapply_dl_model back to the coordinate system of the original image.

Execution Information

Multithreading type: reentrant (runs in parallel with non-exclusive operators).
Multithreading scope: global (may be called from any thread).
Automatically parallelized on internal data level.

Parameters

TransformPipelineTransformPipelineTransformPipelinetransformPipelinetransform_pipeline (input_control) dl_transform_pipeline → (handle)

Transform pipeline handle.

DLSampleBatchDLSampleBatchDLSampleBatchDLSampleBatchdlsample_batch (input_control) dict-array → (handle)

Input data.

DLResultBatchDLResultBatchDLResultBatchDLResultBatchdlresult_batch (output_control) dict-array → (handle)

Result data.

Possible Predecessors

create_dl_transform_pipelinecreate_dl_transform_pipelineCreateDlTransformPipelineCreateDlTransformPipelinecreate_dl_transform_pipeline

Possible Successors

train_dl_model_batchtrain_dl_model_batchTrainDlModelBatchTrainDlModelBatchtrain_dl_model_batch, apply_dl_modelapply_dl_modelApplyDlModelApplyDlModelapply_dl_model

Module

Foundation

Operators