#Object Search

##Getting Started ###Obtaining Dataset The dataset used in the thesis can be found here (~3GB). This dataset needs to be converted to a more suitable format. The following steps need to be undertaken to do this.

###Setting the Paths It is necessary to modify the file +Scripts/setPaths.m for the program to find the dataset. The variable sourceFolder needs to point to the folder where the downloaded file resides. The variable datasetPath needs to point to the folder where the converted files will be saved in.

###Converting the Dataset When the paths are set correctly run following script in Matlab:

Scripts.extractDataset

This script takes some time to complete (~15 minutes). The process is parallelised and can be speeded up by using the command matlabpool open before running the script.

##Computing ###Occurrence Data To obtain the evaluation plots for the occurrence the following two steps are necessary:

1. Learning and evaluating the data:

    Scripts.computeOccurrenceData
   

2. Plotting the result:

    Scripts.plotOccurrenceComparison2Classes
    Scripts.plotOccurrenceComparisonModels
    Scripts.plotOccurrenceStates
    Scripts.plotOccurrenceValueMatrix
   

The first step generates the file tmpOccurrenceData.mat where all computed data is stored. This file can be loaded instead of recomputing the data.

###Location Data Similar as for the occurrence data there are two steps again. It has to be noted that the first step takes a very long time (>3 hours) to complete. The computation speed profits from multiple matlab workers obtained with matlabpool open, the computation speed scales almost linearly with the number of workers used.

1. Learning and evaluating the data:

    Scripts.computeLocationData
   

2. Plotting the results:

    Scripts.plotLocationComparison
    Scripts.plotLocationProbabilityContours
   

##Modifications ###Adapting other Dataset Necessary data is:

• Depth images

• Calibration matrices

• Object bounding boxes with class annotation

To use another dataset with the code it is necessary to create a new DataHandler that inherits from DataHandlers.DataStructure. Further the scripts Scripts.computeOccurrenceData and Scripts.computeLocationData need to be changed to use the new DataHandler instead of the standard DataHandlers.NYUDataStructure. Another possibility is to write a conversion script that uses DataHandlers.NYUDataStructure.addImage to add the new images to the DataHandler.

###Changing Output All files in the +Scripts folder have a number of parameters in the first cell which can be used to produce different plots.