Presentation | 2021-01-21 [Invited Talk] High-Performance Computing for Image Processing Norishige Fukushima, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | As cameras become higher resolution, the demand for high-performance image processing continues to increase. In this paper, we describe a high-performance computing for image processing for this demand. Speeding up image processing, design of algorithms is important. Also, the impact of implementation, such as programming, is much larger than expected. This is because computer architectures have become more complex, and programs that do not handle parallelization, efficient caching, etc. well must be written to fully exploit the performance of the computer. The Halide programming language is a programming language specialized for image processing that supports these types of descriptions. Halide can describe algorithms and computation schedules separately. In terms of the computation scheduling, how to use memory, how to compute, how to compute in parallel, etc., can be changed arbitrarily without changing the algorithm description. The critical concepts in such changes are "parallelism," "locality," and "redundancy." When data access is localized to increase cache efficiency during parallel computation, redundant computation is required to guarantee asynchronous computation. In this paper, we consider high-performance computing, including the trade-off between "approximated computing" and "accuracy" in addition to the above. In this paper, we introduce some research examples that illustrate these concepts, and describe the prospects for highly efficient computation in image processing. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Acceleration / Approximated Computing / Halide / Image Processing |
Paper # | IE2020-35 |
Date of Issue | 2021-01-14 (IE) |
Conference Information | |
Committee | IE |
---|---|
Conference Date | 2021/1/21(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Online |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | Image Processing, Image Coding, etc |
Chair | Hideaki Kimata(NTT) |
Vice Chair | Kazuya Kodama(NII) / Keita Takahashi(Nagoya Univ.) |
Secretary | Kazuya Kodama(KDDI Research) / Keita Takahashi(Nagoya Inst. of Tech.) |
Assistant | Shunsuke Iwamura(NHK) / Shinobu Kudo(NTT) |
Paper Information | |
Registration To | Technical Committee on Image Engineering |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | [Invited Talk] High-Performance Computing for Image Processing |
Sub Title (in English) | |
Keyword(1) | Acceleration |
Keyword(2) | Approximated Computing |
Keyword(3) | Halide |
Keyword(4) | Image Processing |
1st Author's Name | Norishige Fukushima |
1st Author's Affiliation | Nagoya Institute of Technology(NITech) |
Date | 2021-01-21 |
Paper # | IE2020-35 |
Volume (vol) | vol.120 |
Number (no) | IE-329 |
Page | pp.pp.7-7(IE), |
#Pages | 1 |
Date of Issue | 2021-01-14 (IE) |