This repository contains a collection of modules to extract features from images or to perform classification tasks on feature vectors. These modules are meant to be used by other demos to build object recognition pipelines.
At present, the following modules for feature extraction are available:
caffeCoderGIECodersparseCoder
Each of them takes as input an image and outputs its vector representation.
The linearClassifierModule instead implements a linear classifier which can be trained and tested on feature vectors. It is included in this repository because its main usage so far has been on top of a feature extraction module in order to perform image classification, but it can be used on any kind of vectors. While the module is currently in use on our platforms providing good performance, we are working to upgrade it in order to make it faster and more accurate.
While
- YARP
- iCub
- icub-contrib-common
- OpenCV (version < 3.0 is required by
sparseCoder)
are needed by all modules, the following dependencies are required only if you plan to build the corresponding module:
- LIBLINEAR: needed by
linearClassifierModule - SiftGPU: needed by
sparseCoder - Caffe: needed by
caffeCoder - TensorRT: needed by
GIECoder - CUDA and cuDNN: optional for
caffeCoderbut mandatory forGIECoder
Instructions on how to setup the dependencies for each module can be found in specific README files:
caffeCoder: link to READMEGIECoder: link to READMEsparseCoder: link to READMElinearClassifierModule: link to README
Get the code:
$ git clone https://github.com/robotology/himrep.git
And then do, as usual:
$ cd himrep
$ mkdir build && cd build
$ ccmake ../
Where you will configure the project by setting to ON the modules you want to compile and to OFF the ones you want to skip.
IMPORTANT When you run the ccmake command, ensure that:
- the
CMAKE_INSTALL_PREFIXpoints to theicub-contrib-commoninstallation directory - the
YARP_DIR,ICUB_DIR,OpenCV_DIRare correctly pointing to valid installation paths
After that, you can compile and install as usual:
$ make
$ make install
Online autogenerated documentation is available here: https://robotology.github.io/himrep.
Material included here is Copyright of iCub Facility - Istituto Italiano di Tecnologia and is released under the terms of the GPL v2.0 or later. See the file LICENSE for details.