ENGINEERING
Dr. Grandpierre Thierry ESIEE Engineering Cite Descartes
93162 Noisy le Grand Cedex
Tel: 0145926553
Email: t.grandpierre(q)esiee.fr
Mr Jiri Zlatuska
MAS ARYK University
Faculty of Informatics
Department for research & Development
Botanická 68a,
602 00 Brno
November 29, 2010
Obj.: Pavel Karas dissertation report
Dear Mr Jiri Zlatuska,
This report concerns the dissertation thesis topic developed by Pavel Karas. This works is centered on a novel area of computing named GPGPU. GPGPU stands for General Purpose computing using Graphics Processing Units. GPU where initially developed for 3D real time rendering engine but they are now able to be used for accelerating a large set of scientific computations. As state in this document, Pavel Karas aims to use GPGPU in order to accelerate image processing in the area of light microscopy which is the specialty of his lab (Centre for Biomedical Image Analysis - CBIA).
This interesting report is organized in 3 main parts: the first deal with the current results, the second includes the dissertations thesis intent and the third is a summary of study results.
The first part, which is also the most developed part (11 pages), is made of a brief presentation of the latest GPU architectures and is followed by an exhaustive synthesis of bibliography reading (more than 130 references) devoted to image processing on GPU. This bibliography is well structured, and it is evident that Pavel Karas has spent a lot of effort in order to find and present recent articles on each area of image processing (image formation, image enhancement, image analysis...). Each area is briefly but efficiently introduced and organized in pertinent subparts. At this point I would only suggest to develop the architecture chapter as much as the image processing chapter (may be divided it into two independent parts) and I suggest also to add a section dedicated to general purpose multi-core architecture and specialized architectures such for example the IBM CELL.
CITE DESCARTES
2, BOULEVARD BLAISE PASCAL - BP 99 93162 NOISY LE GRAND CEDEX
TEL. : 33 (0) 1 45 92 65 00 FAX.: 33 (0) 1 45 92 66 99
infos@esiee.fr • www.esiee.fr
une école de la
Chambre de commerce et d'industrie de Paris
ESJEE
ENGINEERING
The second part shortly presents the dissertation thesis intent (2 pages). Pavel Karas gives a long list of image processing algorithms that he wishes to implement using GPU in order to accelerate the computing time. The motivating idea seems to be the enhancement of the i3dalgo library developed by the local laboratory. More information about this library may be given to the lecturer. This list is organized in 3 classes: image restoration, image segmentation and morphological processing. Pavel proposes a realist schedule in order to study and implement them. He also takes into account the work he wishes to during a planed internship he will be doing on spring 2011.
In this part, Pavel also states that he will use CUDA programming language and may be "OpenCL if there will be enough time". I would suggest defending more scientifically the choice of the implementation language since it is crucial in several terms: CUDA is a proprietary language developed by Nvidia whereas OpenCL is becoming a standard compatible with Nvidia GPU but also compatible with ATI/AMD GPU and moreover any multicore architecture. So it is important to analyze deeply the impact of the language in term of performance and portability (for the library).
Nevertheless, I agree with the proposed schedule of this second part.
If the first two parts of the dissertation witness of the understanding skill of Pavel Karas, the last part gives evidences of his practical skill. Effectively, he has extended the 3D image library of CBIA laboratory with a GPU implementation of large 3D image convolution algorithm. Moreover, this works has been published in the well known international EURASIP Journal on Advances in Signal Processing. He has also published some encouraging results of the implementation of morphological grayscale reconstruction algorithm onto GPU in a annual doctoral workshop.
Finally, from this dissertation and the two published articles of Pavel Karas there is no doubt that his thesis is well engaged. He has already produce interesting implementation and proposes a good schedule in order to prepare the following of his thesis.
CITE DESCARTES
2, BOULEVARD BLAISE PASCAL - BP < 93162 NOISY LE GRAND CEDEX
TEL.: 33 (0) 1 45 92 65 00 FAX.: 33 (0) 1 45 92 66 99
infos@esiee.fr • www.esiee.fr
une ecole de la
Chambre de commerce et d'industrie de Paris