Classification

Algorithms for classifying object into small, medium, large or predicting class of the object using machine learning models.

Table 1. Available fields in blickfeld.percept_pipeline.config.Classification
Field	Type	Default	Unit
size	Size	-	-
size	Rule-based classification using object size thresholds.
model	Model	-	-
model	Model-based classification using a trained machine learning model.

Size

The minimum size limits of the bounding box area used for classification.

Table 2. Available fields in blickfeld.percept_pipeline.config.Classification.Size
Field	Type	Default	Unit
bounding_box_surface	BoundingBoxSurface	-	-
bounding_box_surface	Use object surface size for object classification.

The maximum surface area between the bounding box’s OXZ and OYZ planes is used to classify objects based on the limits specified in this message.

Table 3. Available fields in blickfeld.percept_pipeline.config.Classification.Size.BoundingBoxSurface
Field	Type	Default	Unit
small_min_size	float	-	m^2
small_min_size	The surface area of the objects bounding box should be larger than this value to be considered a small object.
medium_min_size	float	-	m^2
medium_min_size	The surface area of the objects bounding box should be larger than this value to be considered a medium object.
large_min_size	float	-	m^2
large_min_size	The surface area of the objects bounding box should be larger than this value to be considered a large object.

Configuration for model-based classification.

Table 4. Available fields in blickfeld.percept_pipeline.config.Classification.Model
Field	Type	Default	Unit
weights	bytes	-	-
weights	Serialized model weights.
type	Type	-	-
type	The type of the model.
normalization	Normalization	-	-
normalization	The normalization parameters used to preprocess input features.
feature_generator	FeatureGenerator	-	-
feature_generator	The feature generator method used.
object_category	ObjectClass	-	-
object_category	The subset of classes to use.

Configuration for feature normalization before feeding them into the model.

Table 5. Available fields in blickfeld.percept_pipeline.config.Classification.Model.Normalization
Field	Type	Default	Unit
mean_coeffs	repeated float	-	-
mean_coeffs	The mean coefficients used to normalize each input feature. For each input feature: (x - mean) / std
std_coeffs	repeated float	-	-
std_coeffs	The standard deviation coefficients used for normalization.

Machine learning model types.

Table 6. Available values for blickfeld.percept_pipeline.config.Classification.Model.Type enum
Name	Value	Description
TYPE_UNSPECIFIED	0	The zero value should not be used
TYPE_SVM	1	Support Vector Machine Model

Enum specifying which feature generation method to use for ML model predictions.

Table 7. Available values for blickfeld.percept_pipeline.config.Classification.Model.FeatureGenerator enum
Name	Value	Description
FEATURE_GENERATOR_UNSPECIFIED	0	The zero value should not be used
FEATURE_GENERATOR_EIGVALS_DIMS	1	Generate features from: - Eigenvalues (λ1, λ2, λ3) of the 3D covariance matrix of the object’s point cloud. - Object dimensions: height, width, and length. Resulting feature vector: [λ1, λ2, λ3, height, width, length]
FEATURE_GENERATOR_EIGVALS_DIMS_WITH_VARIANCE	2	Generate features from: - Eigenvalues (λ1, λ2, λ3) of the 3D covariance matrix. - Object dimensions: height, width, and length. - Per-axis variance: var_x, var_y, var_z. Resulting feature vector: [λ1, λ2, λ3, height, width, length, var_x, var_y, var_z]