Implementation of a program to classify objects within of an image. The classification scheme is carried out by two steps. The first stage makes use of a selective search algorithm [5], which selects and proposes classification regions (region proposals) called patches. The algorithm is already included in the OpenCV [2] library. The second stage processes all patches, generated by the first step, with a Convolutional Neural Network based on GoogLeNet [4] a neural network previously trained with Caffe [3]. This part implements a method with message-passing interface (MPI) for sending patches with possibly object candidates to a different nodes containing a neural network trained for classification. The outcome of classification is processed by a primary node, consolidating the different regions of the image.
-
M. P. Forum. Mpi: A message-passing interface standard. Technical report, Knoxville, TN, USA, 1994.
-
Itseez. Open source computer vision library. https://github.com/itseez/opencv, 2015.
-
Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Girshick, S. Guadarrama, and T. Darrell. Caffe: Convolutional architecture for fast feature embedding. arXiv preprint arXiv:1408.5093, 2014.
-
C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabi- novich. Going deeper with convolutions. In Computer Vision and Pattern Recognition (CVPR), 2015.
-
J. Uijlings, K. van de Sande, T. Gevers, and A. Smeulders. Selective search for object recognition. International Journal of Computer Vision, 2013.