Package rst.classification

In machine learning and statistics, classification is the problem of identifying to which of a set of categories (sub-populations) a new observation belongs.

This package contains data types representing classification problems and classification results (e.g. ClassificationResult).

See also

Wikipedia article containing the definition above

http://en.wikipedia.org/wiki/Statistical_classification

Messages

• ClassifiedRegions2D
• ClassifiedRegions3D
• ClassifiedRegion2D
• ClassifiedRegion3D
• ClassificationResult

clearer: should be made invisible via css

Message ClassifiedRegions2D

class rst.classification.ClassifiedRegions2D

A set of image regions with classification results.

Code author: Leon Ziegler <lziegler@techfak.uni-bielefeld.de>

regions

Type :array of rst.classification.ClassifiedRegion2D

The set of regions.

The order of elements is not significant.

Download this file

message ClassifiedRegions2D {

    /**
     * The set of regions.
     *
     * The order of elements is not significant.
     */
    repeated ClassifiedRegion2D regions = 1;

}

Message ClassifiedRegions3D

class rst.classification.ClassifiedRegions3D

A set of 3D regions with classification results.

Code author: Leon Ziegler <lziegler@techfak.uni-bielefeld.de>

regions

Type :array of rst.classification.ClassifiedRegion3D

The set of regions.

The order of elements is not significant.

Download this file

message ClassifiedRegions3D {

    /**
     * The set of regions.
     *
     * The order of elements is not significant.
     */
    repeated ClassifiedRegion3D regions = 1;

}

Message ClassifiedRegion2D

class rst.classification.ClassifiedRegion2D

Focus on image coordinate systems (vision-based).

A image region with a classification result.

Code author: Leon Ziegler <lziegler@techfak.uni-bielefeld.de>

region

Type :rst.geometry.BoundingBox

Region in the input image.

result

Type :rst.classification.ClassificationResult

The class represented by the image region.

Download this file

message ClassifiedRegion2D {

    /**
     * Region in the input image.
     */
    optional geometry.BoundingBox region = 1;

    /**
     * The class represented by the image region.
     */
    optional ClassificationResult result = 2;

}

Message ClassifiedRegion3D

class rst.classification.ClassifiedRegion3D

Focus on image coordinate systems (vision-based).

A region in 3D space with a classification result.

Code author: Leon Ziegler <lziegler@techfak.uni-bielefeld.de>

region

Type :rst.geometry.BoundingBox3DFloat

Region in 3D space.

result

Type :rst.classification.ClassificationResult

The class represented by the 3D region.

Download this file

message ClassifiedRegion3D {

    /**
     * Region in 3D space.
     */
    optional geometry.BoundingBox3DFloat region = 1;

    /**
     * The class represented by the 3D region.
     */
    optional ClassificationResult result = 2;

}

Message ClassificationResult

class rst.classification.ClassificationResult

A class representing the result of a classification process.

An example of such a result could be the output of an SVM-based classifier.

In the simplest form, only the name of a decided class is given. However, it is also possible to give the whole vector of possible classes with their probabilities.

Code author: Johannes Wienke <jwienke@techfak.uni-bielefeld.de>

decided_class

Type :OCTET-VECTOR

A shortcut to get the most likely class easily.

This needs to be an ASCII string.

classes

Type :array of rst.classification.ClassificationResult.ClassWithProbability

The complete vector of classes with their probabilities if suitable.

The results in this vector should of cause not contradict decided_class.

Download this file

message ClassificationResult {

    /**
     * A shortcut to get the most likely class easily.
     *
     * This needs to be an ASCII string.
     */
    required bytes decided_class = 1;

    /**
     * TODO
     */
    message ClassWithProbability {

        /**
         * Name of the class (ASCII string).
         */
        required bytes name = 1;

        /**
         * Some kind of confidence, values are application dependent.
         *
         * However, values summing up to 1 are preferred.
         */
        // @unit(TODO)
        required float confidence = 2;
    }

    /**
     * The complete vector of classes with their probabilities if
     * suitable.
     *
     * The results in this vector should of cause not contradict
     * decided_class.
     */
    repeated ClassWithProbability classes = 2;

}

Message ClassWithProbability

class rst.classification.ClassificationResult.ClassWithProbability

TODO

name

Type :OCTET-VECTOR

Name of the class (ASCII string).

confidence

Type :FLOAT32

Unit: TODO

Some kind of confidence, values are application dependent.

However, values summing up to 1 are preferred.

Download this file

    message ClassWithProbability {

        /**
         * Name of the class (ASCII string).
         */
        required bytes name = 1;

        /**
         * Some kind of confidence, values are application dependent.
         *
         * However, values summing up to 1 are preferred.
         */
        // @unit(TODO)
        required float confidence = 2;
    }