Efficient Computational Scheduling of Box and Gaussian FIR Filtering for CPU Microarchitecture

Fukushima, Norishige; Maeda, Yuji; Kawasaki, Yuki; Nakamura, Masahiro; Tsumura, Tomoaki; Sugimoto, Kenjiro; Kamata, Sei‐ichiro

doi:10.23919/apsipa.2018.8659674

Cited by 20 publications

(9 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, there is still little research on the efficient hardware implementation of SDCT filters, which are FIR filters realized as recursive filters. This research is limited to CPU [21], [58]. The 1-pass 2D filter [58] is highly cache-efficient with fewer synchronizations, although only multichannel data can be vectorized.…”

Section: B Short-time Fourier and Sliding Transformsmentioning

confidence: 99%

SlidingConv: Domain-Specific Description of Sliding Discrete Cosine Transform Convolution for Halide

Kanetaka,

Takagi,

Maeda

et al. 2024

IEEE Access

Self Cite

View full text Add to dashboard Cite

Section: B Short-time Fourier and Sliding Transformsmentioning

confidence: 99%

SlidingConv: Domain-Specific Description of Sliding Discrete Cosine Transform Convolution for Halide

Kanetaka,

Takagi,

Maeda

et al. 2024

IEEE Access

Self Cite

View full text Add to dashboard Cite

“…In existing works on common sub-expression elimination, the process is often described and notated using an array index notation or wiring diagrams (Pasko et al, 1999;Mori et al, 2012;Fukushima et al, 2018). While these notations are sufficient for explanation, they do not make the limitations of the manipulations being used apparent, and so make it difficult to consider how different optimizations could be combined.…”

Section: Related Workmentioning

confidence: 99%

Convolutional kernel function algebra

Stow

2022

Front. Comput. Sci.

View full text Add to dashboard Cite

Many systems for image manipulation, signal analysis, machine learning, and scientific computing make use of discrete convolutional filters that are known before computation begins. These contexts benefit from common sub-expression elimination to reduce the number of calculations required, both multiplications and additions. We present an algebra for describing convolutional kernels and filters at a sufficient level of abstraction to enable intuitive common sub-expression based optimizations through decomposing filters into smaller, repeated, kernels. This enables the creation of an enormous search space of potential implementations of filters via algebraic manipulation. We demonstrate how integral image and sliding window optimizations can be expressed in the context of common sub-expression elimination as well as show the direct use case for this algebra in massively SIMD multiply-free contexts such as in cellular processor arrays. We then show that this algebra is general enough to express and optimize kernels that use non-standard semi-rings to enable shortest path algorithms.

show abstract

“…3b, respectively. For computing SSR, we can use sliding-DCT-based convolution 22,[32][33][34] , which has constant-time property per pixel for convolution. FFT's computational order is O(log N) per pixel, where N is the number of image pixels, while sliding-DCT's one is O(1).…”

Section: Illuminating Saliency Mapmentioning

confidence: 99%

Retinex-Based Light Source Generation for Relighting up Night Photography

Oishi

Fukushima

2022

Preprint

Self Cite

View full text Add to dashboard Cite

The lighting up of buildings is one of the entertainments that make a city more colorful, and photographer sometimes changes the lighting by using photo-editing applications. This paper proposes a method for automatically performing such an application that follows the Retinex theory. The Retinex theory indicates that the complex scenes caught by the human visual system are decomposed into illumination and reflectance components. Many Retinex-based applications take much cost into the decomposition, while we show that the decomposition accuracy is less dependent on the final output for the nightscape relighting. First, we propose an extracting method of a relighting saliency map using Retinex with edge-preserving filtering. Second, we propose a sampling method to specify the lighting area in the limited area. We show that relighting within the limited area can be represented without illumination and reflectance decomposition. We then show that this fact can be exploited to realize this application using a simple Retinex decomposition of the single-scale Retinex with fast edge-preserving filters to represent various lighting effects. Experimental results show that the proposed sampling method is successful in keeping the illuminated points in bright locations and equally spaced.Also, the proposed various diffusion method can enhance night skyline photographs with various expressions. Our code and dataset are available at https://github.com/norishigefukushima/RelightingUpNightPhotography.

show abstract

Efficient Computational Scheduling of Box and Gaussian FIR Filtering for CPU Microarchitecture

Cited by 20 publications

References 32 publications

SlidingConv: Domain-Specific Description of Sliding Discrete Cosine Transform Convolution for Halide

SlidingConv: Domain-Specific Description of Sliding Discrete Cosine Transform Convolution for Halide

Convolutional kernel function algebra

Retinex-Based Light Source Generation for Relighting up Night Photography

Contact Info

Product

Resources

About