The module wraps around classifiers and coders making use of hierarchical image representation which are provided in the himrep repository.

Description

This module is responsible for the communication with coders (i.e. sparseCoder or caffeCoder) and the classifier (i.e. linearClassifierModule) for learning and classify feature vectors. Input features are passed, the output are the scores of the SVM machines.

The commands sent as bottles to the module port /himrepClassifier/rpc are the following:

(notation: [.] identifies a vocab, <.> specifies a double, "." specifies a string)

TRAIN
format: [train]
action: starts the training process by waiting for an image and a bounding box where the object is located.

CLASSIFY
format: [classify]
action: starts the classification process by waiting for an image and a bounding box where the object is located, it returns the output scores in the form "class" "score".

FORGET
format: [forget] "class"
action: forgets the "class", deleting all the feature vectors in the database. If "class"="all" all the classes are forgotten.

BURST
format: [burst] [start | stop]
action: If [start] it starts the training process by waiting for a stream of images and bounding boxes where the object is located. If [stop] it stops the current burst session and automatically trains the new class model.

LIST
format: [list]
reply: "ack" ("name_1" "name_2" ... "name_n")
action: returns the list of known classes names.

CHANGE NAME
format: [chname] "old_name" "new_name"
reply: "ack"/"nack"
action: change the current name of a known object with a new name.

Libraries

YARP libraries.
OpenCV 2.2

Ports Created

/himrepClassifier/rpc receives rpc requests for training and recognition.
/himrepClassifier/img:i receives an input image.
/himrepClassifier/img:o outputs the image to the coder.
/himrepClassifier/classify:rpc RPC port that communicates with the classifier module.
/himrepClassifier/SIFTimg:i reads the image with the extracted local descriptors from the coder.
/himrepClassifier/SIFTimg:o outputs the image with the extracted local descriptors.
/himrepClassifier/scores:i reads the classification scores from the classifier.
/himrepClassifier/features:i reads the hierarchical image representation from the coder.
/himrepClassifier/features:o outpus the hierarchical image representation to the classifier.
/himrepClassifier/opc communication with the Object Property Collector.

Parameters

None.

Tested OS

Linux, Windows 7

Author: Sean Ryan Fanello