Skip to main content

MclassGetResult

BoardSupported
Host SystemYes
V4L2Yes
Clarity UHDYes
Concord PoENo
GenTLYes
GevIQYes
GigE VisionYes
IndioNo
Iris GTXYes
Radient eV-CLYes
Rapixo CLYes
Rapixo CoFYes
Rapixo CXPYes
USB3 VisionYes

Get results from a training or prediction result buffer.

Syntax​

void MclassGetResult(
    AIL_ID    ResultClassId,  //in
    AIL_INT64 LabelOrIndex,   //in
    AIL_INT64 ResultType,     //in
    void *    ResultArrayPtr  //out
)

Description​

This function retrieves results of the specified type from a training or prediction result buffer. Result buffers are allocated using MclassAllocResult. You can either retrieve training results, which are available after calling MclassTrain, or you can retrieve prediction results, which are available after calling MclassPredict.

Parameters​

ResultClassId (in, AIL_ID)​

Specifies the identifier of the classification result buffer from which to retrieve results. You can get results from a CNN (M_TRAIN_CNN_RESULT), object detection (M_TRAIN_DET_RESULT), segmentation (M_TRAIN_SEG_RESULT), anomaly detection (M_TRAIN_ANO_RESULT) or tree ensemble (M_TRAIN_TREE_ENSEMBLE_RESULT) training result buffer, or a CNN (M_PREDICT_CNN_RESULT), object detection (M_PREDICT_DET_RESULT), segmentation (M_PREDICT_SEG_RESULT), anomaly detection (M_PREDICT_ANO_RESULT) tree ensemble (M_PREDICT_TREE_ENSEMBLE_RESULT), or ONNX (M_PREDICT_ONNX_RESULT) prediction result buffer.

LabelOrIndex (in, AIL_INT64)​

Specifies what to retrieve. The values in this table are available for all tasks unless otherwise specified. Set this parameter to one of the following values.

For specifying what to retrieve

ValueDescription
M_DEFAULT
M_CLASS_INDEXSpecifies the index of the class for which to get results. You can get the number of included classes in the result buffer using M_NUMBER_OF_CLASSES.
M_INSTANCE_INDEXSpecifies the index of the instance for which to get results. You can get the number of included instances in the result buffer using M_NUMBER_OF_INSTANCES.

This is only available for object detection result buffers. | | M_OUTPUT_INDEX | Specifies the index of the output of an ONNX model for which to get results.

This is only available for ONNX result buffers. | | M_ALL_INSTANCES | Specifies to retrieve results for all instances.

This is only available for object detection result buffers. | | M_GENERAL (default) | Specifies to retrieve global results from the result buffer. |

ResultType (in, AIL_INT64)​

Specifies the type of result to retrieve.

ResultArrayPtr *(out, void)​

Specifies the address of the array in which to write results.

Parameter Associations​

For training (CNN or segmentation) - retrieving general results​

To retrieve general results from a CNN or segmentation training result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_DEV_DATASET_EPOCH_ACCURACY​

Retrieves an array containing the development dataset accuracy for each epoch. Note that the required array size, which you can retrieve with M_NB_ELEMENTS, is equivalent to the M_MAX_EPOCH (MclassControl) of the training context used.

ValueDescription
0.0 <= Value <= 100.0Specifies the development dataset accuracy for each epoch, as a percentage.

M_DEV_DATASET_EPOCH_ERROR_RATE​

Retrieves an array containing the development dataset error rate for each epoch. Note that the required array size, which you can retrieve with M_NB_ELEMENTS, is equivalent to the M_MAX_EPOCH (MclassControl) of the training context used.

ValueDescription
0.0 <= Value <= 100.0Specifies the development dataset error rate for each epoch, as a percentage.

M_TRAIN_DATASET_EPOCH_ACCURACY​

Retrieves an array containing the training dataset accuracy for each epoch. Note that the required array size, which you can retrieve with M_NB_ELEMENTS, is equivalent to the M_MAX_EPOCH (MclassControl) of the training context used.

ValueDescription
0.0 <= Value <= 100.0Specifies the training dataset accuracy for each epoch, as a percentage.

M_TRAIN_DATASET_EPOCH_ERROR_RATE​

Retrieves an array containing the training dataset error rate for each epoch. Note that the required array size, which you can retrieve with M_NB_ELEMENTS, is equivalent to the M_MAX_EPOCH (MclassControl) of the training context used.

ValueDescription
0.0 <= Value <= 100.0Specifies the training dataset error rate for each epoch, as a percentage.

For training (CNN, object detection, segmentation, or anomaly detection) - retrieving general results​

To retrieve general results from a CNN, object detection, segmentation, or anomaly detection training result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_DEV_DATASET_ERROR_ENTRIES​

Retrieves an array containing the AIL_UUID of each image in the development dataset that produced an error, at least once, during training. For segmentation, UUIDs are also listed for entries that have invalid labels. A label can be invalid if Aurora Imaging Library failed to restore a mask, if the mask size does not match the size of the entry image, if not all the pixels in the image are labeled, or if there is an overlap of pixels between different classes.

M_DEV_DATASET_ERROR_ENTRIES_STATUS​

Retrieves the error status of each entry in the development dataset that produced an error. Use M_DEV_DATASET_ERROR_ENTRIES to get the UUID of the entries that produced an error.

ValueDescription
M_FAILED_TO_RESTORE_GROUND_TRUTH_MASKSpecifies that the ground truth mask entry failed to be restored from the file. This only applies to segmentation tasks.
M_GROUND_TRUTH_MASK_DOES_NOT_EXISTSpecifies that the ground truth mask entry does not exist in the dataset. This only applies to segmentation tasks.
M_GROUND_TRUTH_MASK_INVALID_DIMENSIONSSpecifies that the ground truth entry has invalid dimensions; the height/width of the entry does not match the size of the classifier source layer or the size of the specified target image. This only applies to segmentation tasks.
M_IMAGE_FILE_INVALIDSpecifies that the image file associated with the source entry is not a valid image.
M_IMAGE_FILE_NOT_FOUNDSpecifies that the image file associated with the source entry was not found.
M_INVALID_BUFFER_SIZESpecifies that the entry has invalid dimensions; the height/width of the entry does not match the size of the classifier source layer or the size of the specified target image. This applies to M_PREDICT_ENTRY with a segmentation classifier.
M_INVALID_BUFFER_SIZE_BANDSpecifies that the entry has an invalid number of bands; the number of bands of the entry does not match the number of bands of the classifier source layer or the of the specified target image. This applies to M_PREDICT_ENTRY with a segmentation classifier.
M_INVALID_BUFFER_TYPESpecifies that the entry has an invalid type; the entry type does not match the classifier source layer type or the of the specified target image type.
M_OVERLAPPING_GROUND_TRUTH_PIXELSSpecifies that the ground truth entry contains pixels with overlapping classes. This only applies to segmentation tasks.
M_UNLABELED_GROUND_TRUTH_PIXELSSpecifies that the ground truth entry contains unlabeled pixels. This only applies to segmentation tasks.

M_TRAINABLE_DATASETS​

Retrieves the status of the datasets in train results to determine whether you can perform a complete training using this CNN, object detection, segmentation, or anomaly detection training result buffer with an M_NULL identifier for both datasets of MclassTrain.

ValueDescription
M_DATASETS_STATUS_INVALIDSpecifies that training cannot be done with these datasets.
M_DATASETS_STATUS_MISSING_GTSpecifies that training can be done if missing GT (ground truth) is supported. To support missing GT, enable M_SUPPORT_MISSING_GROUND_TRUTH (MclassControl).
M_DATASETS_STATUS_VALIDSpecifies that training can be done with these datasets.

M_TRAIN_DATASET_ERROR_ENTRIES​

Retrieves an array containing the AIL_UUID of each image in the training dataset that produced an error, at least once, during training. For segmentation, UUIDs are also listed for entries that have invalid labels. A label can be invalid if Aurora Imaging Library failed to restore a mask, if the mask size does not match the size of the entry image, if not all the pixels in the image are labeled, or if there is an overlap of pixels between different classes.

M_TRAIN_DATASET_ERROR_ENTRIES_STATUS​

Retrieves the error status of each entry in the training dataset that produced an error. Use M_TRAIN_DATASET_ERROR_ENTRIES to get the UUID of the entries that produced an error.

ValueDescription
M_FAILED_TO_RESTORE_GROUND_TRUTH_MASKSpecifies that the ground truth mask entry failed to be restored from the file. This only applies to segmentation tasks.
M_GROUND_TRUTH_MASK_DOES_NOT_EXISTSpecifies that the ground truth mask entry does not exist in the dataset. This only applies to segmentation tasks.
M_GROUND_TRUTH_MASK_INVALID_DIMENSIONSSpecifies that the ground truth entry has invalid dimensions; the height/width of the entry does not match the size of the classifier source layer or the size of the specified target image. This only applies to segmentation tasks.
M_IMAGE_FILE_INVALIDSpecifies that the image file associated with the source entry is not a valid image.
M_IMAGE_FILE_NOT_FOUNDSpecifies that the image file associated with the source entry was not found.
M_INVALID_BUFFER_SIZESpecifies that the entry has invalid dimensions; the height/width of the entry does not match the size of the classifier source layer or the size of the specified target image. This applies to M_PREDICT_ENTRY with a segmentation classifier.
M_INVALID_BUFFER_SIZE_BANDSpecifies that the entry has an invalid number of bands; the number of bands of the entry does not match the number of bands of the classifier source layer or the of the specified target image. This applies to M_PREDICT_ENTRY with a segmentation classifier.
M_INVALID_BUFFER_TYPESpecifies that the entry has an invalid type; the entry type does not match the classifier source layer type or the of the specified target image type.
M_OVERLAPPING_GROUND_TRUTH_PIXELSSpecifies that the ground truth entry contains pixels with overlapping classes. This only applies to segmentation tasks.
M_UNLABELED_GROUND_TRUTH_PIXELSSpecifies that the ground truth entry contains unlabeled pixels. This only applies to segmentation tasks.

For training (CNN, object detection, or segmentation) - retrieving general results​

To retrieve general results from a CNN, object detection, or segmentation training result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_LAST_EPOCH_UPDATED_PARAMETERS​

Retrieves the index of the epoch where classifier weights were updated the last time.

ValueDescription
M_INVALIDSpecifies that the parameters were not updated.
Value >= 0Specifies the index of the epoch.

M_MINI_BATCH_PER_EPOCH​

Retrieves the number of mini-batches per epoch.

ValueDescription
Value >= 0Specifies the number of mini-batches per epoch.

M_TRAIN_DATASET_MINI_BATCH_LOSS​

Retrieves an array containing the loss error values for each mini-batch during training. You can use the loss to evaluate the lack of confidence (doubt) associated with the classification during training. Note that the required array size, which you can retrieve with M_NB_ELEMENTS, is equivalent to M_MAX_EPOCH (MclassControl) * M_MINI_BATCH_PER_EPOCH (MclassGetResult) of the training context used.

ValueDescription
0.0 <= Value <= 1.0Specifies the training dataset loss for each mini-batch.

For training (segmentation) - retrieving general results​

To retrieve general results from a segmentation training result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_DEV_DATASET_EPOCH_IOU_MEAN​

Retrieves an array containing the mean IOU (Intersection over Union) metric, as a percentage, of the development dataset using the trained context after each epoch.

ValueDescription
0.0 <= Value <= 100.0Specifies the mean IOU (Intersection over Union) metric of the development dataset, as a percentage.

M_TRAIN_DATASET_EPOCH_IOU_MEAN​

Retrieves an array containing the mean IOU (Intersection over Union) metric, as a percentage, of the training dataset using the trained context after each epoch.

ValueDescription
0.0 <= Value <= 100.0Specifies the mean IOU (Intersection over Union) metric of the training dataset, as a percentage.

For training (segmentation or object detection) - retrieving general results​

To retrieve general results from a segmentation or object detection training result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_DEV_DATASET_EPOCH_LOSS​

Retrieves an array containing the loss metric of the development dataset after the last epoch. Note that the required array size, which you can retrieve with M_NB_ELEMENTS, is equivalent to the M_MAX_EPOCH (MclassControl) of the training context used.

ValueDescription
0.0 <= Value <= 1.0Specifies the loss metric of the development dataset after each epoch.

For training (tree ensemble) - retrieving general results​

To retrieve general results from a tree ensemble training result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL. Note, bagging information is typically unreliable if your training dataset has augmented entries.

M_DEV_DATASET_ACCURACY_AFTER_EACH_TREE​

Retrieves an array containing the accuracies on the development dataset after adding each tree to the ensemble.

M_DEV_DATASET_ERROR_RATE_AFTER_EACH_TREE​

Retrieves an array containing the error rates after on the development dataset after adding each tree to the ensemble.

M_FEATURE_IMPORTANCE​

Retrieves an array containing the importance of each feature when training the classifier. This result can be disabled if M_FEATURE_IMPORTANCE_MODE is set to M_DISABLE.

M_NUMBER_OF_ENTRIES_OUT_OF_BAG​

Retrieves the number of out-of-bag dataset entries.

ValueDescription
Value >= 0Specifies the number of out-of-bag dataset entries.

M_NUMBER_OF_FEATURES​

Retrieves the number of features.

ValueDescription
Value >= 1Specifies the number of features.

M_NUMBER_OF_TREES_TRAINED​

Retrieves the number of trained trees. This result is equivalent to the number of trees in the classifier.

ValueDescription
Value >= 1Specifies the number of trained trees.

M_OUT_OF_BAG_ACCURACY​

Retrieves the out-of-bag error accuracy.

ValueDescription
0.0 <= Value <= 100.0Specifies the out-of-bag accuracy, as a percentage.

M_OUT_OF_BAG_ACCURACY_AFTER_EACH_TREE​

Retrieves the array containing the accuracies on the out-of-bag set after adding each tree to the ensemble.

M_OUT_OF_BAG_CONFUSION_MATRIX​

Retrieves the confusion matrix obtained using the out-of-bag set. This is a square matrix. Aurora Imaging Library returns the matrix values, row-by-row, as an array. The array size is equivalent to the square of the total amount of classes, which is the same M_OUT_OF_BAG_CONFUSION_MATRIX_SIZE_X x M_OUT_OF_BAG_CONFUSION_MATRIX_SIZE_Y, or M_NB_ELEMENTS.

M_OUT_OF_BAG_CONFUSION_MATRIX_SIZE_X​

Retrieves the X-dimension of the confusion matrix. This value is calculated using the out-of-bag set.

ValueDescription
Value >= 1Specifies the X-dimension of the confusion matrix.

M_OUT_OF_BAG_CONFUSION_MATRIX_SIZE_Y​

Retrieves the Y-dimension of the confusion matrix. This value is calculated using the out-of-bag set.

ValueDescription
Value >= 1Specifies the Y-dimension of the confusion matrix.

M_OUT_OF_BAG_ERROR_RATE​

Retrieves the out-of-bag error rate.

ValueDescription
0.0 <= Value <= 100.0Specifies the out-of-bag error rate.

M_OUT_OF_BAG_ERROR_RATE_AFTER_EACH_TREE​

Retrieves the array containing the accuracies on the out-of-bag set after adding each tree to the ensemble.

M_PROXIMITY_MATRIX​

Retrieves the proximity matrix. This is a square matrix. Aurora Imaging Library returns the matrix values, row-by-row, as an array. The array size is equivalent M_PROXIMITY_MATRIX_SIZE_X x M_PROXIMITY_MATRIX_SIZE_Y or M_NB_ELEMENTS. In the proximity matrix, the entry in cell (j, k) is a measure of similarity (or distance) between the entries to which row j and column k correspond. The proximity matrix result can be enabled or disabled using M_COMPUTE_PROXIMITY_MATRIX in the train context.

M_PROXIMITY_MATRIX_SIZE_X​

Retrieves the X-dimension of the proximity matrix.

ValueDescription
Value >= 1Specifies the X-dimension of the proximity matrix.

M_PROXIMITY_MATRIX_SIZE_Y​

Retrieves the Y-dimension of the proximity matrix.

ValueDescription
Value >= 1Specifies the Y-dimension of the proximity matrix.

M_SEED_VALUE​

Retrieves the seed used for the training dataset.

ValueDescription
Value >= 1Specifies the seed used for the training dataset.

M_TRAIN_DATASET_ACCURACY_AFTER_EACH_TREE​

Retrieves the array containing the accuracies on the training dataset after adding each tree to the classifier.

M_TRAIN_DATASET_ERROR_RATE_AFTER_EACH_TREE​

Retrieves the array containing the error rates on the training dataset after adding each tree to the classifier.

M_TREE_DEPTHS_ACHIEVED​

Retrieves the array containing the depth achieved by each tree in the classifier.

M_TREE_NUMBER_OF_LEAF_NODES_ACHIEVED​

Retrieves the array containing the number of leaf nodes for each tree in the classifier.

For training (CNN, segmentation, or tree ensemble) - retrieving general results​

To retrieve general results from a CNN, segmentation or tree ensemble training result buffer (ResultClassId), the ResultType parameter can be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_DEV_DATASET_ACCURACY​

Retrieves the accuracy on the development dataset using the trained context.

ValueDescription
0.0 <= Value <= 100.0Specifies the accuracy on the development dataset, as a percentage.

M_DEV_DATASET_CONFUSION_MATRIX​

Retrieves the confusion matrix obtained using the development dataset. This is a square matrix. Aurora Imaging Library returns the matrix values, row-by-row, as an array. The array size is equivalent to the square of the total amount of classes, which is the same M_DEV_DATASET_CONFUSION_MATRIX_SIZE_X x M_DEV_DATASET_CONFUSION_MATRIX_SIZE_Y, or M_NB_ELEMENTS.

M_DEV_DATASET_CONFUSION_MATRIX_SIZE_X​

Retrieves the X-dimension of the confusion matrix. This value is calculated using the development dataset.

ValueDescription
Value >= 1Specifies the X-dimension of the confusion matrix.

M_DEV_DATASET_CONFUSION_MATRIX_SIZE_Y​

Retrieves the Y-dimension of the confusion matrix. This value is calculated using the development dataset.

ValueDescription
Value >= 1Specifies the Y-dimension of the confusion matrix.

M_DEV_DATASET_ERROR_RATE​

Retrieves the error rate on the development dataset. This can be expressed as 100.0 - M_DEV_DATASET_ACCURACY.

ValueDescription
0.0 <= Value <= 100.0Specifies the accuracy error rate on the development dataset, as a percentage.

M_TRAIN_DATASET_ACCURACY​

Retrieves the accuracy on the training dataset using the trained context.

ValueDescription
0.0 <= Value <= 100.0Specifies the accuracy on the training dataset, as a percentage.

M_TRAIN_DATASET_CONFUSION_MATRIX​

Retrieves the confusion matrix obtained using the training dataset. This is a square matrix. Aurora Imaging Library returns the matrix values, row-by-row, as an array. The array size is equivalent to the square of the total amount of classes, which is the same M_TRAIN_DATASET_CONFUSION_MATRIX_SIZE_X x M_TRAIN_DATASET_CONFUSION_MATRIX_SIZE_Y, or M_NB_ELEMENTS.

M_TRAIN_DATASET_CONFUSION_MATRIX_SIZE_X​

Retrieves the X-dimension of the confusion matrix. This value is calculated using the training dataset.

ValueDescription
Value >= 1Specifies the X-dimension of the confusion matrix.

M_TRAIN_DATASET_CONFUSION_MATRIX_SIZE_Y​

Retrieves the Y-dimension of the confusion matrix. This value is calculated using the training dataset.

ValueDescription
Value >= 1Specifies the Y-dimension of the confusion matrix.

M_TRAIN_DATASET_ERROR_RATE​

Retrieves the error rate on the training dataset. This can be expressed as 100.0 - M_TRAIN_DATASET_ACCURACY.

ValueDescription
0.0 <= Value <= 100.0Specifies the accuracy error rate on the training dataset, as a percentage.

For training (CNN, object detection, segmentation, anomaly detection or tree ensemble) - retrieving general results​

To retrieve general results from a CNN, object detection, segmentation, anomaly detection, or tree ensemble training result buffer (ResultClassId), the ResultType parameter can be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_DEV_DATASET_USED_ENTRIES​

Retrieves an array containing the AIL_UUID of each image in the development dataset used during the training phase.

M_TRAIN_DATASET_USED_ENTRIES​

Retrieves an array containing the AIL_UUID of each image in the training dataset used during the training phase.

For prediction (CNN, object detection, segmentation, or ONNX) - retrieving general results​

To retrieve general results from any prediction result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_CLASSIFIER_PREDEFINED_TYPE​

Retrieves the type of the predefined classifier.

ValueDescription
M_ADNETSpecifies an ADNet classifier context for anomaly detection.
M_CSNET_COLOR_XLSpecifies an extra large CSNet classifier context that is for color images.
M_CSNET_MSpecifies a medium CSNet classifier context.
M_CSNET_MONO_XLSpecifies an extra large CSNet classifier context that is for monochrome images.
M_CSNET_SSpecifies a small CSNet classifier context.
M_CSNET_XLSpecifies an extra large CSNet classifier context.
M_CUSTOMSpecifies a custom classifier context. This is a predefined classifier for image classification, segmentation, or object detection.
M_FCNET_COLOR_XLSpecifies an extra large legacy FCNet classifier context that is for color images.
M_FCNET_MSpecifies a medium legacy FCNet classifier context.
M_FCNET_MONO_XLSpecifies an extra large legacy FCNet classifier context that is for monochrome images.
M_FCNET_SSpecifies a small legacy FCNet classifier context.
M_FCNET_XLSpecifies an extra large legacy FCNet classifier context.
M_ICNET_COLOR_XLSpecifies an extra large ICNet classifier context that is for color images.
M_ICNET_MSpecifies a medium ICNet classifier context.
M_ICNET_MONO_XLSpecifies an extra large ICNet classifier context that is for monochrome images.
M_ICNET_SSpecifies a small ICNet classifier context.
M_ICNET_XLSpecifies an extra large ICNet classifier context.
M_ODNETSpecifies an ODNet classifier context. This is a predefined classifier for object detection.
M_USER_ONNXSpecifies an ONNX classifier context.

For prediction (CNN, segmentation, or anomaly detection) - retrieving general results​

To retrieve general results from a CNN, segmentation, or anomaly detection prediction result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_PREDICTION_SIZE_X​

Retrieves the number of predictions the deep network has made according to the target image size. For image classification, the map size will typically be 1x1 for a target image the same size as the network's source layer, meaning a single prediction for the whole target image. For a larger target image, the map size can be larger, meaning the classifier has convolved the larger target image and made many predictions doing so. For segmentation, the network outputs a prediction for each pixel in the target image, so the map size will be the same size as the target image. For anomaly detection, this shows the output size of the pixel level anomaly detection results.

ValueDescription
Value >= 0Retrieves the number of entries in the dataset.

M_PREDICTION_SIZE_Y​

Retrieves the number of predictions the deep network has made according to the target image size. For image classification, the map size will typically be 1x1 for a target image the same size as the network's source layer, meaning a single prediction for the whole target image. For a larger target image, the map size can be larger, meaning the classifier has convolved the larger target image and made many predictions doing so. For segmentation, the network outputs a prediction for each pixel in the target image, so the map size will be the same size as the target image. For anomaly detection, this shows the output size of the pixel level anomaly detection results.

ValueDescription
Value >= 0Retrieves the number of entries in the dataset.

For prediction (CNN or segmentation) - retrieving general results​

To retrieve general results from a CNN or segmentation prediction result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_BEST_INDEX_IMAGE_TYPE​

Retrieves the image type to provide when calling MclassDraw with M_DRAW_BEST_INDEX_IMAGE or M_DRAW_BEST_INDEX_CONTOUR_IMAGE.

ValueDescription
M_UNSIGNED + 8Specifies that the image type is 8-bit unsigned.
M_UNSIGNED + 16Specifies that the image type is 16-bit unsigned.

M_RECEPTIVE_FIELD_OFFSET_X​

Retrieves the offset along the X-axis, from the top-left corner of the target image, needed to place the first class score at the center of its receptive field.

ValueDescription
Value >= 0Specifies the classification map offset along the X-axis.

M_RECEPTIVE_FIELD_OFFSET_Y​

Retrieves the offset along the Y-axis, from the top-left corner of the target image, needed to place the first class score at the center of its receptive field.

ValueDescription
Value >= 0Specifies the classification map offset along the Y-axis.

M_RECEPTIVE_FIELD_SIZE_X​

Retrieves the size of the receptive field along the X-axis.

ValueDescription
Value >= 0Specifies the size of the receptive field along the X-axis.

M_RECEPTIVE_FIELD_SIZE_Y​

Retrieves the size of the receptive field along the Y-axis.

ValueDescription
Value >= 0Specifies the size of the receptive field along the Y-axis.

M_RECEPTIVE_FIELD_STRIDE_X​

Retrieves the stride spacing the receptive field centers in the target image along the X-axis.

ValueDescription
Value >= 0.0Specifies the classification map scale along the X-axis.

M_RECEPTIVE_FIELD_STRIDE_Y​

Retrieves the stride spacing the receptive field centers in the target image along the Y-axis.

ValueDescription
Value >= 0.0Specifies the classification map scale along the Y-axis.

For prediction (CNN, segmentation, or tree ensemble) - retrieving general results​

To retrieve general results from a CNN, segmentation, or tree ensemble prediction result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL.

M_NUMBER_OF_CLASSES​

Retrieves the number of classes. For CNN or segmentation, this refers to the number of outputs in the network's last layer. For tree ensemble, this refers to the number of tree ensemble in the classifier. This can also be used in a tree ensemble prediction with LabelOrIndex set to M_GENERAL.

ValueDescription
Value >= 0Retrieves the number of entries in the dataset.

For prediction (CNN, object detection, segmentation, or tree ensemble) - retrieving general or specific results​

To retrieve general results from a CNN, object detection, segmentation, or tree ensemble prediction result buffer (ResultClassId), the ResultTypecan be set to one of the following values. In this case, set the LabelOrIndex parameter to M_GENERAL. To retrieve a class or instance specific result from an object detection result buffer, the LabelOrIndex parameter can be set to M_ALL_INSTANCES, M_CLASS_INDEX(), or M_INSTANCE_INDEX().

M_BEST_CLASS_INDEX​

Retrieves the index of the class with the highest score. For image classification, M_BEST_CLASS_INDEX retrieves one index. For segmentation, the array (map) of values retrieved can be arranged in a 2D image, where the size is equal to M_PREDICTION_SIZE_X multiplied by M_PREDICTION_SIZE_Y. Each value is the index of the class having the best score for that X- Y-pixel location. The index of the class begins at 0. For object detection, to retrieve the class index with the highest score for all instances, set the LabelOrIndex parameter to M_ALL_INSTANCES. The array retrieved will have size equal to the number of instances (MclassGetResultwith M_NUMBER_OF_INSTANCES and LabelOrIndex set to M_GENERAL) and will contain the class index with the highest score for each instance. For object detection, to retrieve the class index with the highest score for all instances of a given class, set the LabelOrIndex parameter to M_CLASS_INDEX(). The array retrieved will have size equal to the number of instances (MclassGetResultwith M_NUMBER_OF_INSTANCES and LabelOrIndex set to M_CLASS_INDEX()) and will contain the class index with the highest score for each instance of a given class. For object detection, to retrieve the class index with the highest score for a specific instance, set the LabelOrIndex parameter to M_INSTANCE_INDEX().

ValueDescription
Value >= 0Specifies the index of the class with the highest score.

M_BEST_CLASS_SCORE​

Retrieves the highest class score. For image classification, M_BEST_CLASS_SCORE retrieves one score. For segmentation, the array (map) of values retrieved can be arranged in a 2D image, where the size is equal to M_PREDICTION_SIZE_X multiplied by M_PREDICTION_SIZE_Y. Each value is the best class score for that X- Y-pixel location. For object detection, to retrieve the highest score for all instances, set the LabelOrIndex parameter to M_ALL_INSTANCES. The array retrieved will have size equal to the number of instances (MclassGetResultwith M_NUMBER_OF_INSTANCES and LabelOrIndex set to M_GENERAL) and will contain the highest score for each instance. For object detection, to retrieve the highest score for all instances of a given class, set the LabelOrIndex parameter to M_CLASS_INDEX(). The array retrieved will have size equal to the number of instances (MclassGetResultwith M_NUMBER_OF_INSTANCES and LabelOrIndex set to M_CLASS_INDEX()) and will contain the highest score for each instance of a given class. For object detection, to retrieve the highest score for a specific instance, set the LabelOrIndex parameter to M_INSTANCE_INDEX().

ValueDescription
0.0 <= Value <= 100.0Specifies the best class score.

For prediction (CNN, segmentation, or tree ensemble) - retrieving general or class-specific results​

To retrieve general results from a CNN, segmentation, or tree ensemble prediction result buffer, set the LabelOrIndex parameter to M_GENERAL; to retrieve class-specific results from a CNN or segmentation prediction result buffer, set the LabelOrIndex parameter the index of a specific class (using M_CLASS_INDEX()). In these cases, the ResultType can be set to one of the following values. Note, you cannot receive class-specific results from a tree ensemble result buffer.

M_NUMBER_OF_PREDICTIONS​

Retrieves the total number of class scores. For general results obtained from a CNN or segmentation predict result buffer, the total corresponds to M_PREDICTION_SIZE_X * M_PREDICTION_SIZE_Y * M_NUMBER_OF_CLASSES scores. The first batch of M_PREDICTION_SIZE_X * M_PREDICTION_SIZE_Y scores are for the first class and the subsequent batches are for the remaining classes. In other words, the data is organized planar-wise. For general results obtained from a tree ensemble predict result buffer, the total corresponds to M_NUMBER_OF_CLASSES. For class-specific results (LabelOrIndex set to M_CLASS_INDEX()) obtained from a CNN or segmentation predict result buffer, the total corresponds to M_PREDICTION_SIZE_X * M_PREDICTION_SIZE_Y.

ValueDescription
Value >= 0Retrieves the total number of class scores.

M_PREDICTION_SCORES​

Retrieves an array of all the class scores. The number of returned values is M_NUMBER_OF_PREDICTIONS. For general results obtained from a CNN or segmentation prediction result buffer, the returned values are organized planar-wise in a 3d volume of size M_PREDICTION_SIZE_X *M_PREDICTION_SIZE_Y * M_NUMBER_OF_CLASSES. Note, each band contains the score of one given class. The index in this returned array of the score of class C at pixel X, Y is: index = (C * MapSizeX * MapSizeY) + (Y * MapSizeX) + X. For general results obtained from a tree ensemble prediction result buffer, the number of class scores is equivalent to M_NUMBER_OF_PREDICTIONS. For class-specific results (LabelOrIndex set to M_CLASS_INDEX()) obtained from a CNN or segmentation prediction result buffer, the returned values are organized in a vector that corresponds to a 2D volume of size M_PREDICTION_SIZE_X *M_PREDICTION_SIZE_Y. The index of the score at location X, Y is: index = (Y * [M_PREDICTION_SIZE_X](../../Reference/class/MclassGetResult.md)) + X.

ValueDescription
0.0 <= Value <= 100.0Specifies all of the class scores.

For prediction (object detection) - retrieving general results​

To retrieve general results from an object detection prediction result buffer, set the LabelOrIndex parameter toM_GENERAL. In this case, the ResultType parameter can be set to one of the following values.

M_SCORE_THRESHOLD​

Retrieves the score threshold from the classifier context.

ValueDescription
Value >= 0.0Retrieves the score threshold.

For prediction (object detection) - retrieving general or class-specific results​

To retrieve general or class-specific results from an object detection prediction result buffer, set the LabelOrIndex parameter toM_GENERAL or M_CLASS_INDEX(), respectively. In these cases, the ResultType parameter can be set to one of the following values.

M_NUMBER_OF_INSTANCES​

Retrieves the number of instances.

ValueDescription
Value >= 0Retrieves the number of instances.

For prediction (object detection) - retrieving results for a specific class, a specific instance, or all instances​

To retrieve results for a specific class, a specific instance, or all instances from an object detection prediction result buffer, set the LabelOrIndex parameter to M_CLASS_INDEX(), M_INSTANCE_INDEX(), or M_ALL_INSTANCES. In these cases, you can set the ResultType parameter to one of the following values.

M_BOX_4_CORNERS​

Retrieves an array containing the coordinates of the four corners of the specified bounding boxes. The 8 coordinates of the first box (b1) are grouped, followed by the 8 coordinates of the second box (b2), repeating this process for all specified boxes in the order b1x1, b1y1, b1x2, b1y2, b1x3, b1y3, b1x4, b1y4, b2x1, b2y1, b2x2, b2y2, b2x3, b2y3, b2x4, b2y4, etc. [Image: MclassBox4Corners.png] The size of the array will be equal to the number of specified instances (M_NUMBER_OF_INSTANCES) multiplied by 8.

M_CENTER_X​

Retrieves the X-coordinates of the centers of the specified bounding boxes. When retrieving results for all instances, or a specific class, the X-coordinates are returned in an array. The size of the array can be retrieved with M_NUMBER_OF_INSTANCES. Only one coordinate will be returned if you are retrieving results for a specific instance.

M_CENTER_Y​

Retrieves the Y-coordinates for the centers of the specified bounding boxes. When retrieving results for all instances, or a specific class, the Y-coordinates are returned in an array. The size of the array can be retrieved with M_NUMBER_OF_INSTANCES. Only one coordinate will be returned if you are retrieving results for a specific instance.

M_HEIGHT​

Retrieves the heights of the specified bounding boxes. When retrieving results for all instances, or a specific class, the heights are returned in an array. The size of the array can be retrieved with M_NUMBER_OF_INSTANCES. Only one height will be returned if you are retrieving results for a specific instance.

M_WIDTH​

Retrieves the widths of the specified bounding boxes. When retrieving results for all instances, or a specific class, the widths are returned in an array. The size of the array can be retrieved with M_NUMBER_OF_INSTANCES. Only one coordinate will be returned if you are retrieving results for a specific instance.

For prediction (anomaly detection) - retrieving general results​

To retrieve general results from an anomaly detection prediction result buffer, set the LabelOrIndex parameter toM_GENERAL. In this case, the ResultType parameter can be set to one of the following values.

M_IMAGE_PREDICTED_ANOMALOUS​

Retrieves whether or not the image is considered anomalous.

ValueDescription
M_FALSESpecifies that the image is normal.
M_TRUESpecifies that the image is anomalous.

M_IMAGE_SCORE​

Retrieves the anomaly score of the whole image.

ValueDescription
0.0 <= Value <= 100.0Specifies the anomaly score of the image.

M_MASK_IMAGE​

Retrieves whether or not each pixel in the image is considered anomalous. The array must have a size of M_PREDICTION_SIZE_X x M_PREDICTION_SIZE_Y.

M_PIXEL_SCORES​

Retrieves the anomaly score of each pixel in the image. The array must have a size of M_PREDICTION_SIZE_X x M_PREDICTION_SIZE_Y.

For prediction (ONNX) - retrieving general results​

To retrieve general results from an ONNX prediction result buffer, set the LabelOrIndex parameter toM_GENERAL. In this case, the ResultType parameter can be set to one of the following values.

M_NUMBER_OF_OUTPUTS​

Retrieves the number of outputs of the ONNX model.

ValueDescription
Value >= 1Retrieves the number of outputs.

For prediction (ONNX) - retrieving results for a specific output​

To retrieve general results from an ONNX prediction result buffer, set the LabelOrIndex parameter toM_OUTPUT_INDEX(). In this case, the ResultType parameter can be set to one of the following values.

M_DATA_TYPE​

Retrieves the tensor element type.

ValueDescription
M_FLOAT+16Specifies that the ONNX type is float16. This is not a basic C++ data type thus its representation is platform dependent. The recommended Aurora Imaging Library data type is AIL_UINT16.
M_FLOAT+32Specifies that the ONNX type is float. The recommended Aurora Imaging Library data type is AIL_FLOAT.
M_FLOAT+64Specifies that the ONNX type is double. The recommended Aurora Imaging Library data type is AIL_DOUBLE.
M_SIGNED+8Specifies that the ONNX type is int8. The recommended Aurora Imaging Library data type is AIL_INT8.
M_SIGNED+16Specifies that the ONNX type is int16. The recommended Aurora Imaging Library data type is AIL_INT16.
M_SIGNED+32Specifies that the ONNX type is int32. The recommended Aurora Imaging Library data type is AIL_INT32.
M_SIGNED+64Specifies that the ONNX type is int64. The recommended Aurora Imaging Library data type is AIL_INT64.
M_UNSIGNED+8Specifies that the ONNX type is uint8. The recommended Aurora Imaging Library data type is AIL_UINT8.
M_UNSIGNED+16Specifies that the ONNX type is uint8. The recommended Aurora Imaging Library data type is AIL_UINT8.

M_OUTPUT_DATA​

Retrieves the output tensor data. You can retrieve the number of elements in the tensor with M_NB_ELEMENTS. Apply the corresponding Aurora Imaging Library data type modifier to retrieve the output data in the appropriate format otherwise, elements are converted to AIL_DOUBLE which can be detrimental, especially for float16 and int64 ONNX data types.

M_OUTPUT_RAW​

Retrieves the output tensor raw data. Retrieves the output tensor raw data. You can retrieve the size (in bytes) of the tensor withM_NB_ELEMENTS. This result must be cast using the M_TYPE_AIL_UINT8 modifier.

M_OUTPUT_SHAPE​

Retrieves the output tensor shape array. You can retrieve the number of elements in the array with M_NB_ELEMENTS. The number of elements composing the shape equals the dimensionality (rank) of the tensor. The values of the shape correspond to the lengths of the dimensions of the tensor. The product of all lengths gives the total number of elements in the tensor.

ValueDescription
Value >= 1Retrieves the output tensor shape array.

For training (CNN, object detection, segmentation, anomaly detection, or tree ensemble) or prediction (CNN, object detection, segmentation, anomaly detection, tree ensemble, or ONNX) - retrieving the operation's resulting status​

To retrieve the status of the training or prediction operation from a result buffer (ResultClassId), the ResultTypecan be set to the following value. In this case, set the LabelOrIndex parameter to M_GENERAL. Unless otherwise specified, the following values apply to a CNN, object detection, segmentation, anomaly detection, or tree ensemble training result buffer, or a CNN, object detection, segmentation, anomaly detection, tree ensemble, or ONNX prediction result buffer.

M_STATUS​

Retrieves information regarding the state of the operation.

ValueDescription
M_COMPLETESpecifies that the operation completed successfully.
M_CURRENTLY_PREDICTINGSpecifies that the prediction operation is currently ongoing. You can only get this status if you are retrieving it from another thread. This only applies if you are retrieving results from a CNN, object detection, segmentation, anomaly detection, tree ensemble, or ONNX prediction result buffer (ResultClassId).
M_CURRENTLY_TRAININGSpecifies that the training operation is currently ongoing. You can only get this status if you are retrieving it from another thread. This only applies if you are retrieving results from a CNN, object detection, segmentation, anomaly detection, or tree ensemble training result buffer (ResultClassId).
M_INTERNAL_ERRORSpecifies that an unexpected error occurred during the operation.
M_INVALID_REGION_BUFFERSpecifies that the target image had an invalid region buffer. This only applies if you are retrieving results from a CNN, object detection, segmentation, anomaly detection, tree ensemble, or ONNX prediction result buffer (ResultClassId).
M_NO_VALID_DEVSET_ENTRYSpecifies that the target image had an invalid region buffer.
M_NON_FINITE_VALUE_DETECTEDSpecifies that the training terminated because a non-finite value was detected when the network's parameters were last updated. These values include NaN (Not a Number) and INF (infinity) cases. Such invalid values are often caused by numerically unstable computations that occur during training due to a learning rate that is too high. This only applies if you are retrieving results from a CNN object detection, or segmentation training result buffer (ResultClassId).
M_NOT_ENOUGH_GPU_MEMORYSpecifies that a memory allocation error occurred while training on the GPU. This only applies if you are retrieving results from a CNN, object detection, segmentation, or anomaly detection training result buffer (ResultClassId).
M_NOT_ENOUGH_MEMORYSpecifies that a memory allocation error occurred during the operation.
M_PREDICT_NOT_PERFORMEDSpecifies that the prediction operation was not performed. This is the initial status. This will be the status shown if MclassPredict is terminated early by an Aurora Imaging Library error. This only applies if you are retrieving results from a CNN, object detection, segmentation, anomaly detection, tree ensemble, or ONNX prediction result buffer (ResultClassId).
M_STOPPED_BY_REQUESTSpecifies that the current execution of the operation was explicitly stopped (MclassControl with M_STOP_PREDICT or M_STOP_TRAIN).
M_TIMEOUT_REACHEDSpecifies that the operation ended because the timeout limit was reached (MclassControl with M_TIMEOUT).
M_TRAINING_NOT_PERFORMEDSpecifies that the training operation was not performed. This is the initial status. This only applies if you are retrieving results from a CNN, object detection, segmentation, anomaly detection, or tree ensemble training result buffer (ResultClassId).

Combination Constants — For determining the required number of elements in the array (array size)​

Optional, cannot be used alone.

Usage: You can add one of the following values to the above-mentioned values to determine the required number of elements in the array (array size).

M_NB_ELEMENTS​

Retrieves the required array size (number of elements) to store the returned values.

Combination Constants — For determining whether results are available to be returned​

Optional.

Usage: You can add one of the following values to the above-mentioned values to determine whether a result is available to be returned.

M_AVAILABLE​

Retrieves whether the requested result type is available for retrieval.

ValueDescription
M_FALSESpecifies that the requested result type is not available.
M_TRUESpecifies that the requested result type is available.

Combination Constants — For specifying the data type​

Optional.

Usage: You can add one of the following values to the above-mentioned values to cast the requested results to the required data type.

M_TYPE_AIL_DOUBLE​

Casts the requested results to an AIL_DOUBLE.

M_TYPE_AIL_FLOAT​

Casts the requested results to an AIL_FLOAT.

M_TYPE_AIL_ID​

Casts the requested results to an AIL_ID.

M_TYPE_AIL_INT​

Casts the requested results to an AIL_INT.

M_TYPE_AIL_INT32​

Casts the requested results to an AIL_INT32.

M_TYPE_AIL_INT64​

Casts the requested results to an AIL_INT64.

M_TYPE_AIL_UINT8​

Casts the requested results to an AIL_UINT8.

This is only available if M_COMPUTE_OUT_OF_BAG_RESULTS is set toM_ENABLE and M_BOOTSTRAP is not set toM_DISABLE.

This is a predefined classifier for image classification.

This is a predefined classifier for segmentation.

This is a predefined classifier for image classification or segmentation.

To calculate the center of the receptive field of each result returned forM_PREDICTION_SCORES (Y,X) in the target image, use the following formula: (_X_ * M_RECEPTIVE_FIELD_STRIDE_X + M_RECEPTIVE_FIELD_OFFSET_X; Y * M_RECEPTIVE_FIELD_STRIDE_Y + M_RECEPTIVE_FIELD_OFFSET_Y).

This value is unavailable if your dataset results were produced using MclassPredict with an ONNX classifier (that is, calling MclassInquire with M_CLASSIFIER_PREDEFINED_TYPE must not return M_USER_ONNX).

Copyright © 2026 Zebra Technologies.