Parallelization Strategy Using Lustre and MPI for Face Detection in HPC Cluster: A Case Study

• Autores: Hugo Eduardo Camacho Cruz, Julio Cesar González Mariño, Jesús Humberto Foullon Peña
• Localización: Computación y Sistemas (CyS), ISSN 1405-5546, ISSN-e 2007-9737, Vol. 24, Nº. 1, 2020, págs. 189-199
• Idioma: inglés
• DOI: 10.13053/cys-24-1-3053
• Enlaces
  • Texto completo
• Resumen

  • Abstract The hardware requirements in object detection systems make these applications a challenge in their development given the high consumption of processing and memory they require for their execution. The detection of certain characteristics; in the case of a face, the profile, as well as the lighting, the distances and the numbers of objects are factors that influence the proper functioning and performance of these implementations. This paper presents an alternative to solve part of this problem through a parallelization strategy using Luster and MPI-IO for face detection in the HPC Cluster. We compare Dlib and OpenCV with our alternative based on the Viola-Jones algorithm called Facedetector_MPI. The tests were executed in HPC cluster with 7 nodes (1 metadata server, 2 object storage server and 2 and 4 lustre clients) and we used images since 4 to 148 faces. The results showed an important reduction in the read time of the image file compared with OpenCV of about 50% when the files are bigger to the stripe size(>1MB). Better is the increase obtained in processing around double compared with Dlib in the large images(>2Mpx) without greatly affecting the hit rate in the face detection.

• Referencias bibliográficas
  • Schroff, F.,Kalenichenko, D.,Philbin, J.. (2015). FaceNet: A Unified Embedding for Face Recognition and Clustering. IEEE Computer Society...
  • Taigman, Y.,Yang, M.,Ranzato, M.,Wolf, L.. (2014). DeepFace: Closing the Gap to Human-Level Performance in Face Verification. IEEE Conference...
  • Viola, P.,Jones, M.. (2001). Rapid object detection using a boosted cascade of simple features. IEEE Computer Society Conference on Computer...
  • Rowley, H.A.,Baluja, S.,Kanade, T.. (1998). Neural network-based face detection. IEEE Transactions on Pattern Analysis and Machine Intelligence....
  • King, D.E.. (2009). Dlib-ml: A machine learning toolkit. The Journal of Machine Learning Research. 10. 1755
  • Yu, S.,Jia, S.,Xu, C.. (2017). Convolutional neural networks for hyperspectral image classification. Neurocomputing. 219. 88-98
  • Chang, C.H.,Lu, C.W.,Yang, C.T.,Chang, T.C.. (2016). An approach of performance comparisons with OpenMP and CUDA parallel programming on multicore...
  • Iordan, C.,Jahre, M.,Natvig, L.. (2015). Tuning the victim selection policy of Intel TBB. Journal of Systems Architecture. 61. 58-591
  • Dinan, J.,Balaji, P.,Buntinas, D.,Goodell, D.,Gropp, W.,Thakur, R.. (2016). An implementation and evaluation of the MPI 3.0 onesided communication...
  • Wang, L.,Ma, Y.,Zomaya, A.Y.,Ranjan, R.,Chen, D.. (2015). A parallel file system with application-aware data layout policies for massive remote...
  • Moore, M.,Farrell, P.,Cernohous, B.. (2018). Lustre lockahead: Early experience and performance using optimized locking. Concurrency and Computation:...
  • Pulli, K.,Baksheev, A.,Kornyakov, K.,Eruhimov, V.. (2012). Realtime computer vision with OpenCV. Communications of the ACM. 55. 61-669
  • Crow, F.C.. (1984). Summed-Area tables for texture mapping. ACM SIGGRAPH Computer Graphics. 18. 207