HyperRGBD

[ license-image ] license

Description

C++ framework for building new datasets by aggregating images from different existing RGB-D datasets. The framework is described in:

Petrelli A., Di Stefano L., "Learning to Weight Color And Depth for RGB-D Image Search", International Conference on Image Analysis and Processing, 2017.

Webpage

http://www.vision.deis.unibo.it/research/78-cvlab/107-hyperrgbd

Usage

The integration of an existing dataset into the framework is accomplished by implementing the IDataset interface that requires the implementation of methods:

• ReadImage() loads from disk an image in the required format (e.g. RGB, depth map, 3D point cloud, mask image etc.).
• GetTrainingSet() returns the training set as a list of images, each one denoted by the associated filename and label.
• GetTestSet() returns the test set as a list of images, each one denoted by the associated filename and label.

The aggregation of datasets into new ones is enabled by the HyperDataset class, that requires:

• The definition of the mapping between the categories of the existing datasets and those of the aggregated one through the m_mapCatAssociations map.
• Each aggregated dataset implements a version of the GetTrainingSet() method that returns all the images comprising the dataset.
• The definition of a criteria for splitting the training and test set through the GetTrainingSet() and GetTestSet() methods.

A few examples can be found in hyperrgbdTestMain.cpp

Dependencies

The framework requires OpenCV library for handling images. Moreover, depth maps and calibration data of BigBIRD dataset are stored in HDF5 format.

The code has been tested on Windows 7 and Microsoft Visual Studio 2010.