Operator Reference
get_dl_model_param (Operator)
get_dl_model_param — Return the parameters of a deep learning model.
Signature
get_dl_model_param( : : DLModelHandle, GenParamName : GenParamValue)
Description
get_dl_model_param returns the parameter values
GenParamValue of GenParamName for the deep learning
model DLModelHandle.
For a deep learning model, parameters GenParamName can be set using
set_dl_model_param or create_dl_model_detection, depending
on the parameter and the model type.
With this operator, get_dl_model_param, you can retrieve
the parameter values GenParamValue.
Below we give an overview of the different parameters and an explanation,
except of those you can only set. For latter ones, please see the
documentation of corresponding operator.
The parameters are listed corresponding to the deep learning model methods:
To mark which operators are available for the methods, we use the following notations:
-
set: The parameter can be set usingset_dl_model_param. -
get: The parameter can be retrieved usingget_dl_model_param. -
create: The parameter can be set usingcreate_dl_model_detection. -
Certain parameters are set as non-optional parameters, the corresponding notation is given in brackets.
3D Gripping Point Detection
GenParamName |
set |
get
|
|---|---|---|
| 'adam_beta1' | x |
x
|
| 'adam_beta2' | x |
x
|
| 'adam_epsilon' | x |
x
|
| 'batch_size' | x |
x
|
| 'batch_size_multiplier' | x |
x
|
| 'batchnorm_momentum' | x |
|
| 'class_ids' | x |
x
|
| 'class_names' | x
|
|
| 'device' | x |
x
|
| 'enable_resizing' | x |
|
| 'fuse_bn_relu' | x |
|
| 'fuse_conv_relu' | x |
|
| 'gpu' | x |
x
|
| 'image_dimensions' | x |
x
|
| 'image_height', 'image_width' | x |
x
|
| 'image_num_channels' | x
|
|
| 'image_range_max', 'image_range_min' | x |
x
|
| 'image_size' | x |
x
|
| 'input_dimensions' | x
|
|
| 'learning_rate' | x |
x
|
| 'meta_data' | x |
x
|
| 'min_version' | x
|
|
| 'momentum' | x |
x
|
| 'num_trainable_params' | x
|
|
| 'optimize_for_inference' | x |
x
|
| 'precision' | x
|
|
| 'precision_is_converted' | x
|
|
| 'runtime' | x |
x
|
| 'runtime_init' | x |
|
| 'solver_type' | x |
x
|
| 'type' | x
|
|
| 'weight_prior' | x |
x
|
Anomaly Detection
GenParamName |
set |
get
|
|---|---|---|
| 'batch_size' | x |
x
|
| 'batchnorm_momentum' | x |
|
| 'complexity' | x |
x
|
| 'device' | x |
x
|
| 'enable_resizing' | x |
|
| 'fuse_bn_relu' | x |
|
| 'fuse_conv_relu' | x |
|
| 'gpu' | x |
x
|
| 'image_dimensions' | x |
x
|
| 'image_height', 'image_width' | x |
x
|
| 'image_num_channels' | x
|
|
| 'image_range_max', 'image_range_min' | x
|
|
| 'image_size' | x |
x
|
| 'input_dimensions' | x |
x
|
| 'meta_data' | x |
x
|
| 'min_version' | x
|
|
| 'num_trainable_params' | x
|
|
| 'precision' | x
|
|
| 'precision_is_converted' | x
|
|
| 'runtime' | x |
x
|
| 'runtime_init' | x |
|
| 'standard_deviation_factor' | x |
x
|
| 'type' | x
|
Classification
GenParamName |
set |
get
|
|---|---|---|
| 'adam_beta1' | x |
x
|
| 'adam_beta2' | x |
x
|
| 'adam_epsilon' | x |
x
|
| 'batch_size' | x |
x
|
| 'batch_size_multiplier' | x |
x
|
| 'batchnorm_momentum' | x |
|
| 'class_ids' | x
|
|
| 'class_names' | x |
x
|
| 'class_weights' | x |
x
|
| 'clear_ood' | x |
|
| 'device' | x |
x
|
| 'enable_resizing' | x |
|
| 'extract_feature_maps' | x |
x
|
| 'fuse_bn_relu' | x |
|
| 'fuse_conv_relu' | x |
|
| 'gpu' | x |
x
|
| 'image_dimensions' | x |
x
|
| 'image_height', 'image_width' | x |
x
|
| 'image_num_channels' | x |
x
|
| 'image_range_max', 'image_range_min' | x
|
|
| 'image_size' | x |
x
|
| 'input_dimensions' | x |
x
|
| 'layer_names' | x
|
|
| 'learning_rate' | x |
x
|
| 'meta_data' | x |
x
|
| 'min_version' | x
|
|
| 'momentum' | x |
x
|
| 'num_trainable_params' | x
|
|
| 'ood_available' | x
|
|
| 'ood_threshold' | x |
x
|
| 'optimize_for_inference' | x |
x
|
| 'precision' | x
|
|
| 'precision_is_converted' | x
|
|
| 'runtime' | x |
x
|
| 'runtime_init' | x |
|
| 'solver_type' | x |
x
|
| 'summary' | x
|
|
| 'type' | x
|
|
| 'weight_prior' | x |
x
|
Multi-Label Classification
GenParamName |
set |
get
|
|---|---|---|
| 'batch_size' | x |
x
|
| 'batch_size_multiplier' | x |
x
|
| 'class_ids' | x |
x
|
| 'class_names' | x |
x
|
| 'device' | x |
x
|
| 'gpu' | x |
x
|
| 'image_dimensions' | x |
x
|
| 'image_height', 'image_width' | x |
x
|
| 'image_num_channels' | x |
x
|
| 'image_range_max', 'image_range_min' | x
|
|
| 'image_size' | x |
x
|
| 'input_dimensions' | x |
x
|
| 'layer_names' | x
|
|
| 'learning_rate' | x |
x
|
| 'meta_data' | x |
x
|
| 'min_version' | x
|
|
| 'min_confidence' | x |
x
|
| 'momentum' | x |
x
|
| 'num_trainable_params' | x
|
|
| 'optimize_for_inference' | x |
x
|
| 'precision' | x
|
|
| 'precision_is_converted' | x
|
|
| 'runtime' | x |
x
|
| 'runtime_init' | x |
|
| 'solver_type' | x |
x
|
| 'summary' | x
|
|
| 'type' | x
|
|
| 'weight_prior' | x |
x
|
Deep Counting component
GenParamName |
set |
get
|
|---|---|---|
| 'batch_size' | x |
x
|
| 'batch_size_multiplier' | x |
x
|
| 'device' | x |
x
|
| 'gpu' | x |
x
|
| 'input_dimensions' | x
|
|
| 'layer_names' | x
|
|
| 'learning_rate' | x |
x
|
| 'meta_data' | x |
x
|
| 'min_version' | x
|
|
| 'momentum' | x |
x
|
| 'num_trainable_params' | x
|
|
| 'optimize_for_inference' | x
|
|
| 'precision' | x
|
|
| 'precision_is_converted' | x
|
|
| 'runtime' | x |
x
|
| 'solver_type' | x
|
|
| 'summary' | x
|
|
| 'type' | x
|
|
| 'weight_prior' | x |
x
|
Deep OCR detection component
GenParamName |
set |
get
|
|---|---|---|
| 'adam_beta1' | x |
x
|
| 'adam_beta2' | x |
x
|
| 'adam_epsilon' | x |
x
|
| 'batch_size' | x |
x
|
| 'batchnorm_momentum' | x |
|
| 'device' | x |
x
|
| 'enable_resizing' | x |
|
| 'fuse_bn_relu' | x |
|
| 'fuse_conv_relu' | x |
|
| 'gpu' | x |
x
|
| 'image_dimensions' | x |
x
|
| 'image_height', 'image_width' | x |
x
|
| 'image_num_channels' | x |
x
|
| 'image_range_max', 'image_range_min' | x |
x
|
| 'image_size' | x |
x
|
| 'input_dimensions' | x |
x
|
| 'layer_names' | x
|
|
| 'learning_rate' | x |
x
|
| 'meta_data' | x |
x
|
| 'min_character_score' | x |
x
|
| 'min_link_score' | x |
x
|
| 'min_version' | x
|
|
| 'min_word_area' | x |
x
|
| 'min_word_score' | x |
x
|
| 'momentum' | x |
x
|
| 'num_trainable_params' | x
|
|
| 'orientation' | x |
x
|
| 'optimize_for_inference' | x |
x
|
| 'precision' | x
|
|
| 'precision_is_converted' | x
|
|
| 'runtime' | x |
x
|
| 'runtime_init' | x |
|
| 'solver_type' | x |
x
|
| 'sort_by_line' | x |
x
|
| 'summary' | x
|
|
| 'tiling' | x |
x
|
| 'tiling_overlap' | x |
x
|
| 'type' | x
|
|
| 'weight_prior' | x |
x
|
Deep OCR recognition component
GenParamName |
set |
get
|
|---|---|---|
| 'adam_beta1' | x |
x
|
| 'adam_beta2' | x |
x
|
| 'adam_epsilon' | x |
x
|
| 'alphabet' | x |
x
|
| 'alphabet_internal' | x |
x
|
| 'alphabet_mapping' | x |
x
|
| 'batch_size' | x |
x
|
| 'batchnorm_momentum' | x |
|
| 'device' | x |
x
|
| 'enable_resizing' | x |
|
| 'fuse_bn_relu' | x |
|
| 'fuse_conv_relu' | x |
|
| 'gpu' | x |
x
|
| 'image_dimensions' | x |
x
|
| 'image_height', 'image_width' | x |
x
|
| 'image_num_channels' | x |
x
|
| 'image_range_max', 'image_range_min' | x |
x
|
| 'image_size' | x |
x
|
| 'input_dimensions' | x |
x
|
| 'layer_names' | x
|
|
| 'learning_rate' | x |
x
|
| 'meta_data' | x |
x
|
| 'min_version' | x
|
|
| 'momentum' | x |
x
|
| 'num_trainable_params' | x
|
|
| 'optimize_for_inference' | x |
x
|
| 'precision' | x
|
|
| 'precision_is_converted' | x
|
|
| 'runtime' | x |
x
|
| 'runtime_init' | x |
|
| 'solver_type' | x |
x
|
| 'summary' | x
|
|
| 'type' | x
|
|
| 'weight_prior' | x |
x
|
Global Context Anomaly Detection
GenParamName |
set |
get
|
|---|---|---|
| 'adam_beta1' | x |
x
|
| 'adam_beta2' | x |
x
|
| 'adam_epsilon' | x |
x
|
| 'anomaly_score_tolerance' | x |
x
|
| 'batch_size' | x |
x
|
| 'batch_size_multiplier' | x |
x
|
| 'batchnorm_momentum' | x |
|
| 'device' | x |
x
|
| 'enable_resizing' | x |
|
| 'fuse_bn_relu' | x |
|
| 'fuse_conv_relu' | x |
|
| 'gc_anomaly_networks' | x |
x
|
| 'gpu' | x |
x
|
| 'image_dimensions' | x |
x
|
| 'image_height', 'image_width' | x |
x
|
| 'image_num_channels' | x |
x
|
| 'image_range_max', 'image_range_min' | x
|
|
| 'image_size' | x |
x
|
| 'input_dimensions' | x
|
|
| 'layer_names' | x
|
|
| 'learning_rate' | x |
x
|
| 'meta_data' | x |
x
|
| 'min_version' | x
|
|
| 'momentum' | x |
x
|
| 'num_trainable_params' | x
|
|
| 'optimize_for_inference' | x |
x
|
| 'patch_size' | x |
x
|
| 'precision' | x
|
|
| 'precision_is_converted' | x
|
|
| 'runtime' | x |
x
|
| 'runtime_init' | x |
|
| 'solver_type' | x |
x
|
| 'summary' | x
|
|
| 'type' | x
|
|
| 'weight_prior' | x |
x
|
Object Detection, Instance Segmentation
GenParamName |
set |
get |
create
|
|---|---|---|---|
| 'adam_beta1' | x |
x |
|
| 'adam_beta2' | x |
x |
|
| 'adam_epsilon' | x |
x |
|
| 'anchor_angles' | x |
x
|
|
| 'anchor_aspect_ratios' | x |
x
|
|
| 'anchor_num_subscales' | x |
x
|
|
'backbone' (Backbone) |
x |
x
|
|
| 'backbone_docking_layers' | x |
x |
x
|
| 'batch_size' | x |
x |
|
| 'batch_size_multiplier' | x |
x |
|
| 'batchnorm_momentum' | x |
||
| 'bbox_heads_weight', 'class_heads_weight' | x |
x |
x
|
| 'capacity' | x |
x
|
|
| 'class_ids' | x |
x |
x
|
| 'class_ids_no_orientation' | x |
x
|
|
| 'class_names' | x |
x |
x
|
| 'class_weights' | x |
||
| 'device' | x |
x |
|
| 'enable_resizing' | x |
||
| 'freeze_backbone_level' | x |
x |
x
|
| 'fuse_bn_relu' | x |
||
| 'fuse_conv_relu' | x |
||
| 'gpu' | x |
x |
|
| 'ignore_direction' | x |
x
|
|
| 'image_dimensions' | x |
x
|
|
| 'image_height', 'image_width' | x |
x
|
|
| 'image_num_channels' | x |
x
|
|
| 'image_range_max', 'image_range_min' | x |
||
| 'image_size' | x |
x
|
|
| 'input_dimensions' | x |
||
| 'instance_segmentation' | x |
x
|
|
| 'instance_type' | x |
x
|
|
| 'layer_names' | x |
||
| 'learning_rate' | x |
x |
|
| 'mask_head_weight' | x |
x |
x
|
| 'max_level', 'min_level' | x |
x
|
|
| 'max_num_detections' | x |
x |
x
|
| 'max_overlap' | x |
x |
x
|
| 'max_overlap_class_agnostic' | x |
x |
x
|
| 'meta_data' | x |
x |
|
| 'min_confidence' | x |
x |
x
|
| 'min_version' | x |
||
|---|---|---|---|
| 'momentum' | x |
x |
|
'num_classes' (NumClasses) |
x |
x
|
|
| 'num_trainable_params' | x |
||
| 'optimize_for_inference' | x |
x |
x
|
| 'precision' | x |
||
| 'precision_is_converted' | x |
||
| 'runtime' | x |
x |
|
| 'runtime_init' | x |
||
| 'solver_type' | x |
x |
|
| 'summary' | x |
||
| 'type' | x |
||
| 'weight_prior' | x |
x |
Semantic Segmentation, Edge Extraction
GenParamName |
set |
get
|
|---|---|---|
| 'adam_beta1' | x |
x
|
| 'adam_beta2' | x |
x
|
| 'adam_epsilon' | x |
x
|
| 'batch_size' | x |
x
|
| 'batch_size_multiplier' | x |
x
|
| 'batchnorm_momentum' | x |
|
| 'class_ids' | x |
x
|
| 'class_names' | x |
x
|
| 'device' | x |
x
|
| 'enable_resizing' | x |
|
| 'fuse_bn_relu' | x |
|
| 'fuse_conv_relu' | x |
|
| 'gpu' | x |
x
|
| 'ignore_class_ids' | x |
x
|
| 'image_dimensions' | x |
x
|
| 'image_height', 'image_width' | x |
x
|
| 'image_num_channels' | x |
x
|
| 'image_range_max', 'image_range_min' | x |
x
|
| 'image_size' | x |
x
|
| 'input_dimensions' | x |
x
|
| 'layer_names' | x
|
|
| 'learning_rate' | x |
x
|
| 'meta_data' | x |
x
|
| 'min_version' | x
|
|
| 'momentum' | x |
x
|
| 'num_classes' | x
|
|
| 'num_trainable_params' | x
|
|
| 'optimize_for_inference' | x |
x
|
| 'precision' | x
|
|
| 'precision_is_converted' | x
|
|
| 'runtime' | x |
x
|
| 'runtime_init' | x |
|
| 'solver_type' | x |
x
|
| 'summary' | x
|
|
| 'type' | x
|
|
| 'weight_prior' | x |
x
|
In the following we list and explain the parameters GenParamName
for which you can retrieve their value using this operator,
get_dl_model_param.
Note, that only parameters that do not change the model architecture can be
set after the model has been optimized with
optimize_dl_model_for_inference. A list of these parameters can be
found at optimize_dl_model_for_inference.
Thereby, the following symbols denote the model type for which the parameter
can be get:
-
'Any': any method
-
'3D-GPD': 'type'='3d_gripping_point_detection'
-
'AD': 'type'='anomaly_detection'
-
'CL': 'type'='classification'
-
'MLC': 'type'='multi_label_classification'
-
'DC': 'type'='counting'
-
'OCR-D': 'type'='ocr_detection'
-
'OCR-R': 'type'='ocr_recognition'
-
'GC-AD': 'type'='gc_anomaly_detection'
-
'OD': 'type'='detection'
-
'SE': 'type'='segmentation'
-
'Gen': 'type'='generic'
- 'adam_beta1': 3D-GPD CL OCR-D OCR-R GC-AD OD SE Gen
-
This value defines the moment for the linear term in Adam solver. For more information we refer to the documentation of
train_dl_model_batch.Restriction: Only applicable for 'solver_type' = 'adam'.
Default: 'adam_beta1' = 0.9
- 'adam_beta2': 3D-GPD CL OCR-D OCR-R GC-AD OD SE Gen
-
This value defines the moment for the quadratic term in Adam solver. For more information we refer to the documentation of
train_dl_model_batch.Restriction: Only applicable for 'solver_type' = 'adam'.
Default: 'adam_beta2' = 0.999
- 'adam_epsilon': 3D-GPD CL OCR-D OCR-R GC-AD OD SE Gen
-
This value defines the epsilon in the Adam solver formula and is purposed to guarantee the numeric stability. For more information we refer to the documentation of
train_dl_model_batch.Restriction: Only applicable for 'solver_type' = 'adam'.
Default: 'adam_epsilon' = 1e-08
- 'alphabet': OCR-R
-
The character set that can be recognized by the Deep OCR model.
It contains all characters that are not mapped to the Blank character of the internal alphabet (see parameters 'alphabet_mapping' and 'alphabet_internal').
The alphabet can be changed or extended if needed. Changing the alphabet with this parameter will edit the internal alphabet and mapping in such a way that it tries to keep the length of the internal alphabet unchanged. After changing the alphabet, it is recommended to retrain the model on application specific data (see the HDevelop example
deep_ocr_recognition_training_workflow.hdev). Previously unknown characters will need more training data.Note, that if the length of the internal alphabet changes, the last model layers have to be randomly initialized and thus the output of the model will be random strings (see 'alphabet_internal'). In that case it is required to retrain the model.
- 'alphabet_internal': OCR-R
-
The full character set which the Deep OCR recognition component has been trained on.
The first character of the internal alphabet is a special character. In the pretrained model this character is specified as Blank (U+2800) and is not to be confused with a space. The Blank is never returned in a word output but can occur in the reported character candidates. It is required and cannot be omitted. If the internal alphabet is changed, the first character has to be the Blank. Furthermore, if 'alphabet' is used to change the alphabet, the Blank symbol is added automatically to the character set.
The length of this tuple corresponds to the depth of the last convolution layer in the model. If the length changes, the last convolution layer and all layers after it have to be resized and potentially reinitialized randomly. After such a change, it is required to retrain the model (see HDevelop example
deep_ocr_recognition_training_workflow.hdev).It is recommended to use the parameter 'alphabet' to change the alphabet, as it will automatically try to preserve the length of the internal alphabet.
- 'alphabet_mapping': OCR-R
-
Tuple of integer indices.
It is a mapping that is applied by the model during the decoding step of each word. The mapping overwrites a character of 'alphabet_internal' with the character at the specified index in 'alphabet_internal'.
In the decoding step each prediction is mapped according to the index specified in this tuple. The tuple has to be of same length as the tuple 'alphabet_internal'. Each integer index of the mapping has to be within 0 and |'alphabet_internal'|-1.
In some applications it can be helpful to map certain characters onto other characters. E.g. if only numeric words occur in an application it might be helpful to map the character "O" to the "0" character without the need to retrain the model.
If an entry contains a 0, the corresponding character in 'alphabet_internal' will not be decoded in the word.
- 'anchor_angles': OD
-
The parameter 'anchor_angles' determines the orientation angle of the anchors for a model of 'instance_type' = 'rectangle2'.
Thereby, the orientation is given in arc measure and indicates the angle between the horizontal axis and
Length1(mathematically positive). See the chapter Deep Learning / Object Detection and Instance Segmentation for more explanations to anchors.You can set a tuple of values. A higher number of angles increases the number of anchors which might lead to a better localization but also increases the runtime and memory-consumption.
Value range: 'anchor_angles' for 'ignore_direction' = 'false', 'anchor_angles' for 'ignore_direction' = 'true'
Default: 'anchor_angles' = [0.0]
- 'anchor_aspect_ratios' (legacy: 'aspect_ratios'): OD
-
The parameter 'anchor_aspect_ratios' determines the aspect ratio of the anchors. Thereby, the definition of the ratio depends on the 'instance_type':
-
'rectangle1': height-to-width ratio
-
'rectangle2': ratio length1 to length2
E.g., for instance type 'rectangle1' the ratio 2 gives a narrow and 0.5 a broad anchor. The size of the anchor is affected by the parameter 'anchor_num_subscales' and with its explanation we give the formula for the sizes and lengths of the generated anchors. See the chapter Deep Learning / Object Detection and Instance Segmentation for more explanations to anchors.
You can set a tuple of values. A higher number of aspect ratios increases the number of anchors which might lead to a better localization but also increases the runtime and memory-consumption.
For reasons of backward compatibility, the parameter name 'aspect_ratios' can be used instead of 'anchor_aspect_ratios'.
Default: 'anchor_aspect_ratios' = [1.0, 2.0, 0.5]
-
- 'anchor_num_subscales' (legacy: 'num_subscales'): OD
-
This parameter determines the number of different sizes with which the anchors are generated at the different levels used.
In HALCON for every anchor point, thus every pixel of every feature map of the feature pyramid, a set of anchors is proposed. See the chapter Deep Learning / Object Detection and Instance Segmentation for more explanations to anchors. Thereby the parameter 'anchor_num_subscales' affects the size of the anchors. An example is shown in the figure below.
With 'anchor_num_subscales'=2 we generate for every aspect ratio 2 anchors of different size on each level: One with the base length (solid line) and an additional, larger one (dotted line). Thereby, in the image these additional anchors of the lower level (orange) converge to the anchor of the next higher level (blue).An anchor of level has by default a edge lengths of in the input image, whereby the parameter has the value . With the parameter 'anchor_num_subscales' additional anchors can be generated, which converge in size to the smallest anchor of the level . More precisely, these anchors of level have in the input image the edge lengths where . For subscale , this results on level in an anchor of height and width equal where is the ratio of this anchor (see 'anchor_aspect_ratios').
A larger number of subscales increases the number of anchors and will therefore increase the runtime and memory-consumption.
For reasons of backward compatibility, the parameter name 'num_subscales' can be used instead of 'anchor_num_subscales'.
Default: 'anchor_num_subscales' = 3
- 'anomaly_score_tolerance': GC-AD
-
The image-level anomaly score is calculated internally such that a certain fraction of the pixel-level anomaly scores in
anomaly_imageis greater or equal to the image-level anomaly score. The value of this fraction can be set by 'anomaly_score_tolerance' and has to be in the interval [0.0, 1.0]. For example, for 'anomaly_score_tolerance'=0.01, 1 percent of the pixel anomaly scores are larger or equal to the image anomaly score. This can be used to suppress outliers that might appear in the pixel anomaly scores.Default: 'anomaly_score_tolerance' = 0.0
- 'backbone': OD
-
The parameter 'backbone' is the name (together with the path) of the backbone network which is used to create the model. A list of the delivered backbone networks can be found under
create_dl_model_detection. - 'backbone_docking_layers': CL
-
The parameter 'backbone_docking_layers' specifies which layers of the backbone are to be used as docking layers by the feature pyramid. Thereby the layers are referenced by their names.
The docking layers can be specified for every classifier, also without using them as backbone. The specification is only considered for object detection backbones. When selecting the docking layers, consider that the feature map lengths have to be halved from one docking layer to the other. Rule of thumb: Use the deepest layers for every (lateral) resolution in the backbone (corresponding to one of the required levels for your object detection task).
Information about the names and sizes of the layers in a model can be enquired using 'summary'.
Default: For the pretrained backbones delivered by HALCON the defaults depend on the classifier. Other classifiers do not have any docking layers set by default and therefore need to have this parameter set before they can be used as backbone.
- 'batch_size': Any
-
Number of input images (and corresponding labels) in a batch that is transferred to device memory.
For a training using
train_dl_model_batch, the batch of images which are processed simultaneously in a single training iteration contains a number of images which is equal to 'batch_size' times 'batch_size_multiplier'. Please refer totrain_dl_model_batchfor further details.For inference, the 'batch_size' can be generally set independently from the number of input images. See
apply_dl_modelfor details on how to set this parameter for greater efficiency.- Models of 'type'='classification':
-
The parameter 'batch_size' is stored in the pretrained classifier. Per default, the 'batch_size' is set such that a training of the pretrained classifier with up to
For the pretrained classifiers, the default values are hence given as follows:100classes can be easily performed on a device with8gigabyte of memory.pretrained classifier default value of 'batch_size' 'pretrained_dl_classifier_alexnet.hdl' 230 'pretrained_dl_classifier_compact.hdl' 160 'pretrained_dl_classifier_enhanced.hdl' 96 'pretrained_dl_classifier_mobilenet_v2.hdl' 40 'pretrained_dl_classifier_resnet18.hdl' 24 'pretrained_dl_classifier_resnet50.hdl' 23
- Models of 'type'='counting':
The parameter 'batch_size' has no effect.
- 'batch_size_multiplier': Any
-
Multiplier for 'batch_size' to enable training with larger numbers of images in one step which would otherwise not be possible due to GPU memory limitations. This model parameter does only affect
train_dl_model_batchand thus has no impact during evaluation and inference. For detailed information seetrain_dl_model_batch.- Models of 'type'='anomaly_detection':
The parameter 'batch_size_multiplier' has no effect.
- Models of 'type'='counting':
The parameter 'batch_size_multiplier' has no effect.
- Models of 'type'='ocr_recognition':
The parameter 'batch_size_multiplier' has no effect.
Default: 'batch_size_multiplier' = 1
- 'bbox_heads_weight', 'class_heads_weight': OD
-
The parameters 'bbox_heads_weight' and 'class_heads_weight' are weighting factors for the calculation of the total loss. This means, when the losses of the individual networks are summed up, the contributions from the bounding box regression heads are weighted by a factor 'bbox_heads_weight' and the contributions from the classification heads are weighted by a factor 'class_heads_weight'.
Default: 'bbox_heads_weight' = 1.0, 'class_heads_weight' = 1.0
- 'capacity': OD
-
This parameter roughly determines the number of parameters (or filter weights) in the deeper sections of the object detection network (after the backbone). Its possible values are 'high', 'medium', and 'low'.
It can be used to trade-off between detection performance and speed. For simpler object detection tasks, the 'low' or 'medium' settings may be sufficient to achieve the same detection performance as with 'high'.
Default: 'capacity' = 'high'
- 'class_ids': 3D-GPD CL MLC OD SE
-
Unique IDs of the classes the model shall distinguish. The tuple is of length 'num_classes'.
We stress out the slightly different meanings and restrictions depending on the model type:
- Models of 'type'='3d_gripping_point_detection':
-
Two classes, in fixed order ['gripping_map', 'background'], are supported for this model. Therefore, for such a model the tuple has a fixed length of 2. Thereby, you can set any integer within the interval as class ID value.
- Models of 'type'='classification':
-
The IDs are unique identifiers, which are automatically assigned to each class. The ID of a class corresponds to the index within the tuple 'class_names'.
- Models of 'type'='multi_label_classification':
-
The IDs are unique identifiers of the classes to be detected. Thereby, you can set any integer as class ID value.
Value range: .
- Models of 'type'='detection':
-
Only the classes of the objects to be detected are included and therewith no background class. Thereby, you can set any integer within the interval as class ID value.
Note that the values of 'class_ids_no_orientation' depend on 'class_ids'. Thus if 'class_ids' is changed after the creation of the model, 'class_ids_no_orientation' is reset to an empty tuple.
Default: 'class_ids' = '[0,...,num_classes-1]'
- Models of 'type'='segmentation':
Every class used for training has to be included and therewith also the class ID of the 'background' class. Therefore, for such a model the tuple has a minimal length of 2. Thereby, you can set any integer within the interval as class ID value.
- 'class_ids_no_orientation': OD
-
With this parameter you can declare classes, for which the orientation will not be considered, e.g., round or other point symmetrical objects. For each class, whose class ID is present in 'class_ids_no_orientation', the network returns axis-aligned bounding boxes.
Note, this parameter only affects networks of 'instance_type' = 'rectangle2'.
Note that the values of 'class_ids_no_orientation' depend on 'class_ids'. Thus if 'class_ids' is changed after the creation of the model, 'class_ids_no_orientation' is reset to an empty tuple.
Default: 'class_ids_no_orientation' = []
- 'class_names': 3D-GPD CL MLC OD SE
-
Unique names of the classes the model shall distinguish. The order of the class names remains unchanged after the setting. The tuple is of length 'num_classes'.
Restriction: For a '3d_gripping_point_detection' model, the 'class_names' tuple has a fixed length of 2, with constant value ['gripping_map', 'background'].
- 'class_weights': CL OD
-
The parameter 'class_weights' is a tuple of class specific weighting factors for the loss. Giving the unique classes a different weight, it is possible to force the network to learn the classes with different importance. This is useful in cases where a class dominates the dataset. The weighting factors have to be within the interval . Thereby a class gets a stronger impact during the training the larger its weight is. The weights in the tuple 'class_weights' are sorted the same way as the classes in the tuple 'class_ids'. We stress out the slightly different meanings and restrictions depending on the model type:
- Models of 'type'='classification':
Default: 'class_weights' = 1.0 for each class.
- Models of 'type'='detection':
Default: 'class_weights' = 0.25 for each class.
- 'clear_ood': CL
-
This parameter deactivates the out-of-distribution detection of a deep learning model. All related information will be deleted.
fit_dl_out_of_distributioncan be called to fit the out-of-distribution detection again.Default: 'true'
- 'complexity': AD
-
This parameter controls the capacity of the model to deal with more complex applications. A higher value allows for the model to represent images showing more complexity. Increasing the parameter leads to higher runtimes during training and inference. Please note that this parameter can only be set before the model is trained. Setting 'complexity' on an already trained model would render this model useless. When trying to do so, an error is returned but the model itself is unchanged.
Default: 'complexity' = 15
- 'device': Any
-
Handle of the device on which the deep learning operators will be executed.
If the model was already optimized for a device, setting 'device' might not be necessary anymore, see
optimize_dl_model_for_inferencefor details.To get a tuple of handles of all available potentially deep-learning capable hardware devices use
query_available_dl_devices.Default: Handle of the default device, thus the GPU with index 0. If not available, this is an empty tuple.
- 'extract_feature_maps': CL
-
With this parameter value you can extract feature maps of the specified model layer for an inferred image. The selected layer must be part of the existing network. An overview of all existing layers of the model can be returned by the operator
get_dl_model_paramwith the corresponding parameter 'summary'.Note, using this option modifies the network architecture: The network is truncated after a selected layer. This modification can not be reversed. If the original network architecture should be used again it must be read in again with the operator
read_dl_model. - 'freeze_backbone_level': OD
-
This parameter determines the backbone levels whose weights are kept (meaning not updated and thus frozen) during training. Thereby the given number signifies the highest level whose layers are frozen in the backbone. Setting 'freeze_backbone_level' to 0, for no level the weights are frozen and as a consequence the weights of all layers are updated. It is recommended to set this in case the weights have been randomly initialized (e.g., after certain changes of the number of image channels) or the in case the backbone is not pretrained.
Default: 'freeze_backbone_level' = 2
- 'gc_anomaly_networks': GC-AD
-
This parameter is used to select the subnetworks of the model. The following values can be set:
-
'local': The local subnetwork will be extracted.
-
'global': The global subnetwork will be extracted.
We refer to Deep Learning / Anomaly Detection and Global Context Anomaly Detection for more general information on the subnetworks and their properties.
Note, the original model contains both a 'local' and a 'global' subnetwork. Once the model architecture is reduced to a single subnetwork, the original network is truncated. This cannot be undone. If the original model architecture is to be used again, it must be read in again with
read_dl_model.Default: 'gc_anomaly_networks' = ['local', 'global']
-
- 'gpu': Any
-
Identifier of the GPU where the training and inference operators (
train_dl_model_batchandapply_dl_model) are executed. Per default, the first available GPU is used.get_systemwith 'cuda_devices' can be used to retrieve a list of available GPUs. Pass the index in this list to 'gpu'.Note that the parameter 'gpu' is only taken into account for 'runtime' = 'gpu'. Therefore, it is preferable to set the GPU device, on which operators are run, using the parameter 'device'. executed.
Default: 'gpu' = 0
- 'ignore_class_ids': SE
-
With this parameter you can declare one or multiple classes as 'ignore' classes, see the chapter Deep Learning / Semantic Segmentation and Edge Extraction for further information. These classes are declared over their ID (integers).
Note, you can not set a class ID in 'ignore_class_ids' and 'class_ids' simultaneously.
- 'ignore_direction': OD
-
This parameter determines whether for the oriented bounding box also the direction of the object within the bounding box is considered or not. In case the direction within the bounding box is not to be considered you can set 'ignore_direction' to 'true'. In order to determine the bounding box unambiguously, in this case (but only in this case) the following conventions apply:
-
'phi'
-
'bbox_length1' > 'bbox_length2'
This is consistent to
smallest_rectangle2.Note, this parameter only affects networks of 'instance_type' = 'rectangle2'.
List of values: 'true', 'false'
Default: 'ignore_direction' = 'false'
-
- 'image_dimensions': 3D-GPD AD CL MLC OCR-D OCR-R GC-AD OD SE
-
Tuple containing the input image dimensions 'image_width', 'image_height', and number of channels 'image_num_channels'.
The respective default values and possible value ranges depend on the model and model type. Please see the individual dimension parameter description for more details.
- 'image_height', 'image_width': 3D-GPD AD CL MLC OCR-D OCR-R GC-AD OD SE
-
Height and width of the input images, respectively, that the network will process.
This parameter can attain different values depending on the model type:
- Models of 'type'='3d_gripping_point_detection':
-
The network architectures allow changes of the image height and width.
The default values are given by the network, see
read_dl_model. - Models of 'type'='anomaly_detection':
-
The default values depend on the specific pretrained network, see
read_dl_model. The network architectures allow changes of the image dimensions, which can be done usingset_dl_model_param. Please also refer toread_dl_modelfor restrictions each of the delivered networks has on the input image size. Note that these parameters have to be set before training the model. Setting them on an already trained model would render this model useless. When trying to do so, an error is returned but the model itself is unchanged. - Models of 'type'='classification':
-
The default values depend on the specific pretrained classifier, see
read_dl_model. The network architectures allow changes of the image dimensions, which can be done usingset_dl_model_param. But for networks with at least one fully connected layer such a change makes a retraining necessary. Networks without fully connected layers are directly applicable to different image sizes. However, images with a size differing from the size with which the classifier has been trained are likely to show a reduced classification accuracy. - Models of 'type'='detection':
-
The network architectures allow changes of the image dimensions. But the image lengths are halved for every level, that is why the dimensions 'image_width' and 'image_height' need to be an integer multiple of . depends on the 'backbone' and the parameter 'max_level', see
create_dl_model_detectionfor further information.Default: 'image_height' = 640, 'image_width' = 640
- Models of 'type'='gc_anomaly_detection':
-
The network architecture allows changes of the image dimensions. Note that these parameters have to be set before training the model. Setting them on an already trained model would render this model useless.
Restriction: The 'image_width' and the 'image_height' must be greater than or equal to the value of the 'patch_size' parameter.
Default: 'image_height' = 256, 'image_width' = 256
- Models of 'type'='ocr_recognition':
-
The network architectures allow changes of the image width.
The default and minimal values are given by the network, see
read_dl_model. - Models of 'type'='segmentation':
-
The network architectures allow changes of the image dimensions.
The default and minimal values are given by the network, see
read_dl_model.
- 'image_num_channels': 3D-GPD AD CL MLC OCR-D OCR-R GC-AD OD SE
-
Number of channels of the input images the network will process. The default value is given by the network, see
read_dl_modelandcreate_dl_model_detection.For models of 'type'='anomaly_detection' or 'type'='gc_anomaly_detection', only the values 1 and 3 are supported. In addition, this parameter should be set before the model is trained. For models of 'type'='anomaly_detection', setting 'image_num_channels' on an already trained model would render this model useless. Therefore, an error is returned when trying to do so, but the model itself is unchanged.
Restriction: For models of 'type'='3d_gripping_point_detection' the parameter 'image_num_channels' cannot be set.
For other models, any number of input image channels is possible.
If number of channels is changed to a value >1, the weights of the first layers after the input image layer will be initialized with random values. Note, in this case more data for the retraining is needed. If the number of channels is changed to 1, the weights of the concerned layers are fused.
- Models of 'type'='anomaly_detection':
Default: 'image_num_channels' = 3
- Models of 'type'='detection':
Default: 'image_num_channels' = 3
- 'image_range_max', 'image_range_min': 3D-GPD AD CL MLC OCR-D OCR-R GC-AD OD SE
-
Maximum and minimum gray value of the input images, respectively, the network will process.
The default values are given by the network, see
read_dl_modelandcreate_dl_model_detection. - 'image_size': 3D-GPD AD CL MLC OCR-D OCR-R GC-AD OD SE
-
Tuple containing the input image size 'image_width', 'image_height'.
The respective default values and possible value ranges depend on the model and model type. Please see the individual dimension parameter description for more details.
- 'input_dimensions': Any
-
This parameter returns a dictionary containing all input dimensions of the network. Examples for such inputs: input image,
weight_image(for models of 'type'='segmentation').These dimensions are given in the dictionary as a tuple [
width,height,depth]. In case this parameter is used to set the dimension, for every dimension a value of -1 may be set to keep the current value. - 'instance_segmentation': OD
-
This parameter determines if the model is created for instance segmentation. If the parameter is set to 'true' in
create_dl_model_detection, the detection deep learning network is extended by additional layers for instance segmentation.List of values: 'true', 'false'
Default: 'instance_segmentation' = 'false'
- 'instance_type': OD
-
The parameter 'instance_type' determines, which instance type is used for the object model. The current implementations differ regarding the allowed orientations of the bounding boxes. See the chapter Deep Learning / Object Detection and Instance Segmentation for more explanations to the different types and their bounding boxes.
List of values: 'rectangle1', 'rectangle2'
Default: 'instance_type' = 'rectangle1'
- 'layer_names': Any except 3D-GPD
-
This parameter returns a tuple containing the name for every layer of the model. This name is the same human-readable identifier as is returned by
get_dl_model_paramwith 'summary'.Note, for some networks distributed with HALCON, the network architecture is confidential. In this case
get_dl_model_paramreturns an empty tuple with 'layer_names'. - 'learning_rate': Any
-
Value of the factor determining the gradient influence during training using
train_dl_model_batch. Please refer totrain_dl_model_batchfor further details.The default values depend on the model. Note that changing 'solver_type' sets 'learning_rate' back to its default value.
- Models of 'type'='anomaly_detection':
The parameter 'learning_rate' has no effect.
- Models of 'type'='counting':
The parameter 'learning_rate' has no effect.
- 'mask_head_weight': OD
-
The parameter 'mask_head_weight' is a weighting factor for the calculation of the total loss. This means, when the losses of the individual network heads are summed up, the contribution from the mask prediction head is weighted by a factor 'mask_head_weight'.
Restriction: Only applicable to models with 'instance_segmentation'='true'
Default: 'mask_head_weight' = 1.0
- 'max_level', 'min_level': OD
-
These parameters determine on which levels the additional networks are attached on the feature pyramid. We refer to the chapter Deep Learning / Object Detection and Instance Segmentation for further explanations to the feature pyramid and the attached networks.
From these ('max_level' - 'min_level' + 1) networks all predictions with a minimum confidence value are kept as long they do not strongly overlap (see 'min_confidence' and 'max_overlap').
The level declares how often the size of the feature map already has been scaled down. Thus, level 0 corresponds to the feature maps with size of the input image, level 1 to feature maps subscaled once, and so on. As a consequence, smaller objects are detected in the lower levels, whereas larger objects are detected in higher levels.
The value for 'min_level' needs to be at least 2.
If 'max_level' is larger than the number of levels the backbone can provide, the backbone is extended with additional (randomly initialized) convolutional layers in order to generate deeper levels. Further, 'max_level' may have an influence on the minimal input image size.
Note, for small input image dimensions, high levels might not be meaningful, as the feature maps could already be too small to contain meaningful information.
A higher number of used levels might increase the runtime and memory-consumption, whereby especially lower levels carry weight.
Default: 'max_level' = 6, 'min_level' = 2
- 'max_num_detections': OD
-
This parameter determines the maximum number of detections (bounding boxes) per image proposed from the network.
Default: 'max_num_detections' = 100
- 'max_overlap': OD
-
The maximum allowed intersection over union (IoU) for two predicted bounding boxes of the same class. Or, vice-versa, when two bounding boxes are classified into the same class and have an IoU higher than 'max_overlap', the one with lower confidence value gets suppressed. We refer to the chapter Deep Learning / Object Detection and Instance Segmentation for further explanations to the IoU.
Default: 'max_overlap' = 0.5
- 'max_overlap_class_agnostic': OD
-
The maximum allowed intersection over union (IoU) for two predicted bounding boxes independently of their predicted classes. Or, vice-versa, when two bounding boxes have an IoU higher than 'max_overlap_class_agnostic', the one with lower confidence value gets suppressed. As default, 'max_overlap_class_agnostic' is set to 1.0, hence class agnostic bounding box suppression has no influence.
Default: 'max_overlap_class_agnostic' = 1.0
- 'meta_data': Any
-
Dictionary with user defined meta data, whose entries can be set freely. The meta data may be used to store information such as the model author or a model version along with the model.
Restriction: Dictionary values are limited to strings.
- 'min_character_score': OCR-D
-
The parameter 'min_character_score' specifies the lower threshold used for the character score map to estimate the dimensions of the characters. By adjusting the parameter, suggested instances can be split up or neighboring instances can be merged.
Value range: .
Default: 0.5
- 'min_confidence': MLC OD
- Models of 'type'='detection':
-
This parameter determines the minimum confidence, when the image part within the bounding box is classified in order to keep the proposed bounding box. This means, when
apply_dl_modelis called, all output bounding boxes with a confidence value smaller than 'min_confidence' are suppressed. - Models of 'type'='multi_label_classification':
-
This parameter determines the minimum confidence level required for a class to be considered as detected. This means, classes with a confidence score below 'min_confidence' are regarded as not detected, while classes with a confidence value equal to or exceeding 'min_confidence' are deemed as detected.
Default: 'min_confidence' = 0.5
- 'min_link_score': OCR-D
-
The parameter 'min_link_score' defines the minimum link score required between two localized characters to recognize these characters as coherent word.
Value range: .
Default: 0.3
- 'min_version': Any
-
The parameter 'min_version' defines the minimum required HALCON version to use this model.
- 'min_word_area': OCR-D
-
The parameter 'min_word_area' defines the minimum size that a localized word must have in order to be suggested. This parameter can be used to filter suggestions that are too small.
Value range: .
Default: 10.
- 'min_word_score': OCR-D
-
The parameter 'min_word_score' defines the minimum score a localized instance must contain to be suggested as valid word. With this parameter uncertain words can be filtered out.
Value range: .
Default: 0.7
- 'momentum': Any
-
When updating the weights of the network, the hyperparameter 'momentum' specifies to which extent previous updating vectors will be added to the current updating vector. Only applicable for 'solver_type' = 'sgd'.
Please refer to
train_dl_model_batchfor further details.The default value is given by the model.
- Models of 'type'='anomaly_detection':
The parameter 'momentum' has no effect.
- Models of 'type'='counting':
The parameter 'momentum' has no effect.
- 'num_classes': OD SE
-
Number of distinct classes that the model is able to distinguish for its predictions.
This parameter differs between the model types:
- Models of 'type'='detection':
This parameter is set as
NumClassesovercreate_dl_model_detection. 'class_ids' and 'class_names' always need to have 'num_classes' entries.- Models of 'type'='segmentation':
A model of 'type'='segmentation' does predict background and therefore in this case the 'background' class is included in 'num_classes'. For these models, 'num_classes' is determined implicitly by the length of 'class_ids'.
- 'num_trainable_params': Any
-
Number of trainable parameters (weights and biases) of the model. This value is an indicator for the size of the model when it is serialized.
- 'ood_available': CL
-
This parameter specifies whether a classification model has been extended for Out-of-Distribution Detection. If the operator
fit_dl_out_of_distributionhas been called for the model, the value of 'ood_available' is 'true'. Otherwise it is 'false'.Default: 'false'
- 'ood_threshold': CL
-
A threshold that determines when a sample is classified as Out-of-Distribution. After calling
fit_dl_out_of_distributionfor a classification model,apply_dl_modeladditionally returns an 'ood_score' and an 'ood_result' in the output dictionaries. If the value of 'ood_score' is greater than or equal to 'ood_threshold', the value of 'ood_result' is 'true'. Otherwise it is 'false'. 'ood_threshold' is only available afterfit_dl_out_of_distributionhas been called. That operator calculates a value for 'ood_threshold' based on the provided dataset.Value range: .
- 'orientation': OCR-D
-
This parameter allows to set a predefined orientation angle for the word detection. To revert to default behavior using the internal orientation estimation, 'orientation' is set to 'auto'.
Value range: .
Default: 'auto'
- 'optimize_for_inference': 3D-GPD CL MLC OCR-D OCR-R GC-AD OD SE Gen
-
Defines whether the model is optimized and only applicable for inference.
The model remains executable on HALCON standard devices even after optimization (unlike
optimize_dl_model_for_inference).Setting this parameter to 'true' frees model memory not needed for inference (e.g., memory of gradients). This can significantly reduce the amount of memory needed by the model. As a consequence, models with this characteristic have no gradients accessible (needed e.g., for training or calculations of heatmaps). Operators using values from freed memory (e.g.,
train_dl_model_batch) will automatically reset this parameter value to 'false'.In case the value is reset to 'false' (both, manually or automatically), memory needed by the model for training is reallocated. This implies, a following training behaves as if 'momentum' is temporarily set to 0 (as possible updating vectors have to be accumulated again).
Default: 'false'
Restriction: For models of 'type'='counting' the parameter 'optimize_for_inference' cannot be set to 'true'.
- 'patch_size': GC-AD
-
This parameter determines the size of the patches analyzed by the 'local' subnetwork. The patch size should be chosen such that patches containing defects will differ clearly from patches without defects. This should cover all kinds of anomalies that might occur during inference. The patch size does not need to cover the defects as a whole. Note that the image is not divided into separate patches, but 'patch_size' solely determines the scale on which the image is gradually analyzed by the 'local' subnetwork. If the image size is changed 'patch_size' should be adjusted accordingly.
( 1) ( 2) (1) Examples of patches with a suitable size (green). (2) The example patches (red) are either too large or too small to work for all possible defect classes that could occur in the images.Restriction: The parameter 'patch_size' must be smaller than or equal to 'image_width' and 'image_height'.
Default: 'patch_size' = 33
- 'precision': Any
-
Defines the data type that is internally used for the calculation of a forward pass of a deep learning model.
Default: 'float32'
- 'precision_is_converted': Any
-
Indicates whether the model was subjected to a conversion process after training done by
optimize_dl_model_for_inference.Default: 'false'
- 'runtime': Any
-
Defines the device on which the deep learning operators will be executed.
Note that the parameter 'device' should be preferred to set the devices on which the deep learning operators will be executed.
Default: 'runtime' = 'gpu'
- 'cpu':
-
The training and inference operator will be executed on CPU. Note, training is only supported for specific platform types, please see the HALCON
“Installation Guide”.In case the GPU has been used before, CPU memory is initialized, and if necessary values stored on the GPU memory are moved to the CPU memory.
For parallelization: The runtime is highly dependent on the number of threads set. The use of all available threads does not necessarily create a faster performance. How many threads are currently set can be queried with the operator
get_system.The implemented CPU parallelization is dependent on the architecture:
-
Intel or AMD architecture: OpenMP. By default all available threads of the OpenMP runtime environment are used. The number of threads used can be specified with the parameter 'tsp_thread_num' of the operator
set_system. -
Arm architectures: Global Thread Pool. The number of threads can be set with the global parameter 'thread_num' of the operator
set_system.
For both architectures mentioned above, it is not possible to specify a thread-specific number of threads (via the parameter 'tsp_thread_num' of the operator
set_system). -
- 'gpu':
-
The GPU memory is initialized. The operators
apply_dl_model,train_dl_model_batch, andtrain_dl_model_anomaly_datasetwill be executed on the GPU. For the specific requirements please refer to the HALCON“Installation Guide”.
- 'solver_type': Any
-
This value defines the optimization algorithm with the goal to minimize the value of the total loss function. For more information we refer to the documentation of
train_dl_model_batch. The following values can be set:-
'adam': Adaptive moment estimation
-
'sgd': Stochastic gradient descent
Note that changing 'solver_type' sets 'learning_rate' back to its default value.
- Models of 'type'='anomaly_detection':
The parameter 'solver_type' has no effect.
- Models of 'type'='counting':
The parameter 'solver_type' has no effect.
-
- 'sort_by_line': OCR-D
-
The words are sorted line-wise based on the orientation of the localized word instances. If the parameter 'sort_by_line' is set to 'false', the results will not be sorted.
Default: 'true'
- 'standard_deviation_factor': AD
-
The anomaly score is calculated internally as the mean of certain internal scores s plus lambda times their standard deviation.
Where s denotes a pixel value of the internal
anomaly_image, the mean value of s and the standard deviation of s. The parameter 'standard_deviation_factor' sets the value and thus controls how important the standard deviation is in comparison to the mean.Default: 'standard_deviation_factor' = 3.0
- 'summary': Any
-
This parameter returns information on the layers of the model. More precisely, it returns a tuple with a string for every layer. The string is as follows:
ID; NAME; TYPE; OUTPUT_SHAPE; CONNECTED_NODES-
ID: Index of the layer in the CNN graph. -
NAME: Human-readable identifier (optional). -
TYPE: Human-readable identifier representing the type of the layer (e.g.,inputorconvolution). -
OUTPUT_SHAPE: Size of the output, given in the form (Width,Height,Depth, 'batch_size'). This means, the layer has feature maps of sizeWidthtimesHeightand therefromDepthmany. Together they form an iconic object with a channel for every feature map. The parameter 'batch_size' determines, how many objects are returned together. -
CONNECTED_NODES: Comma separated list with IDs of the layers using the output of the current layer as input
E.g.,
'3; conv1; convolution; (112, 112, 64, 160); 4'.Note, for some networks distributed with HALCON, the network architecture is confidential. In this case
get_dl_model_paramreturns an empty tuple with 'summary'. -
- 'tiling': OCR-D
-
The input image is automatically split into overlapping tile images of size 'image_size', which are processed separately by the model. This allows processing images that are much larger than the actual 'image_size' without having to zoom the input image. Default: 'false'
- 'tiling_overlap': OCR-D
-
This parameter defines how much neighboring tiles overlap when input images are split (see 'tiling'). The overlap is given in pixels.
Value range: .
Default: 64
- 'type': Any
-
This parameter returns the HALCON-specific model type. The following types are distinguished:
-
'3d_gripping_point_detection'
-
'anomaly_detection'
-
'classification'
-
'counting'
-
'detection'
-
'gc_anomaly_detection'
-
'segmentation'
-
'ocr_recognition'
-
'ocr_detection'
-
'generic' - for certain read in models or models created with the DL framework, see
set_dl_model_param.
-
- 'weight_prior': Any
-
Regularization parameter used for the regularization of the loss function. For a detailed description of the regularization term we refer to
train_dl_model_batch. Simply put: Regularization favors simpler models that are less likely to learn noise in the data and generalize better. In case the classifier overfits the data, it is strongly recommended to try different values for the parameter 'weight_prior' to improve the generalization properties of the neural network. Choosing its value is a trade-off between the models ability to generalize, overfitting, and underfitting. If is too small, the model might overfit, if its too large the model might loose its ability to fit the data, because all weights are effectively zero. For finding an ideal value for , we recommend a cross-validation, i.e. to perform the training for a range of values and choose the value that results in the best validation error. For typical applications, we recommend testing the values for 'weight_prior' on a logarithmic scale between . If the training takes a very long time, one might consider performing the hyperparameter optimization on a reduced amount of data.- Models of 'type'='anomaly_detection':
The parameter 'weight_prior' has no effect.
- Models of 'type'='counting':
The parameter 'weight_prior' has no effect.
Default: 'weight_prior' = 0.0, with exception of the pretrained classifiers
-
pretrained_dl_classifier_resnet18: 'weight_prior' = 0.0001 -
pretrained_dl_classifier_resnet50: 'weight_prior' = 0.0001 -
pretrained_dl_classifier_alexnet: 'weight_prior' = 0.0005
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
DLModelHandle (input_control) dl_model → (handle)
Handle of the deep learning model.
GenParamName (input_control) attribute.name → (string)
Name of the generic parameter.
Default: 'batch_size'
List of values: 'adam_beta1', 'adam_beta2', 'adam_epsilon', 'alphabet', 'alphabet_internal', 'alphabet_mapping', 'anchor_angles', 'anchor_aspect_ratios', 'anchor_num_subscales', 'anomaly_score_tolerance', 'backbone', 'backbone_docking_layers', 'batch_size', 'batch_size_multiplier', 'bbox_heads_weight', 'capacity', 'class_heads_weight', 'class_ids', 'class_ids_no_orientation', 'class_names', 'class_weights', 'complexity', 'device', 'extract_feature_maps', 'freeze_backbone_level', 'gc_anomaly_networks', 'gpu', 'ignore_class_ids', 'ignore_direction', 'image_dimensions', 'image_height', 'image_num_channels', 'image_range_max', 'image_range_min', 'image_size', 'image_width', 'input_dimensions', 'instance_segmentation', 'instance_type', 'layer_names', 'learning_rate', 'mask_head_weight', 'max_level', 'max_num_detections', 'max_overlap', 'max_overlap_class_agnostic', 'meta_data', 'min_character_score', 'min_confidence', 'min_level', 'min_link_score', 'min_version', 'min_word_area', 'min_word_score', 'model_uid', 'momentum', 'num_classes', 'num_trainable_params', 'ood_available', 'ood_threshold', 'optimize_for_inference', 'orientation', 'patch_size', 'precision', 'precision_is_converted', 'release_version', 'runtime', 'solver_type', 'sort_by_line', 'standard_deviation_factor', 'summary', 'tiling', 'tiling_overlap', 'type', 'weight_prior'
GenParamValue (output_control) attribute.name(-array) → (integer / string / real)
Value of the generic parameter.
Result
If the parameters are valid, the operator get_dl_model_param
returns the value 2 (
H_MSG_TRUE)
. If necessary, an exception is raised.
Possible Predecessors
read_dl_model,
set_dl_model_param
Possible Successors
set_dl_model_param,
apply_dl_model,
train_dl_model_batch,
train_dl_model_anomaly_dataset
See also
Module
Foundation. This operator uses dynamic licensing (see the 'Installation Guide'). Which of the following modules is required depends on the specific usage of the operator:
3D Metrology, OCR/OCV, Matching, Deep Learning Enhanced, Deep Learning Professional