Vectorized implementation of K-means

Otsuka, Tomoki; Fukushima, Norishige

doi:10.1117/12.2590842

Cited by 4 publications

(2 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, writing high-speed image processing programs requires advanced knowledge and experience with image processing and hardware. There are various tasks in image processing to optimize these code, such as filtering, [1][2][3] wavelet, 4 and K-means; 5 however, these codes tends to be complex.…”

Section: Introductionmentioning

confidence: 99%

Performance evaluation of halide auto-scheduler with directional cubic convolution interpolation

Nogami

Oishi

Sasaki

et al. 2023

International Workshop on Advanced Imaging Technology (IWAIT) 2023

View full text Add to dashboard Cite

Finding the optimal implementation of calculations is one of the most critical challenges in image processing programming. Halide is a domain-specific language for high-performance image processing. Its auto-scheduler is a helpful tool for solving this problem; however, its scheduling is not a panacea for complex flows. In this paper, we evaluate the performance of the auto-scheduler by comparing it to hand-manually implemented C++ codes. The algorithm used for comparison is Directional cubic convolution interpolation (DCCI), whose computation schedule is challenging to optimize. We evaluate three auto-schedulers: Adams et al. 's, Li et al.'s, and Mullapudi et al.s. Experimental results show that the performance of the schedule generated by Adams' method is comparable to that of the hand-implemented C++ code.

show abstract

Section: Introductionmentioning

confidence: 99%

Performance evaluation of halide auto-scheduler with directional cubic convolution interpolation

Nogami

Oishi

Sasaki

et al. 2023

International Workshop on Advanced Imaging Technology (IWAIT) 2023

View full text Add to dashboard Cite

show abstract

“…For example, wavelet, 1 finite impulse response filtering, [2][3][4] infinite impulse response filtering, 5 recursive filtering, 6,7 local Laplacian filtering, 8 bilateral filtering, [9][10][11] frequency filtering, 12,13 guided image filtering, 14 and K-means clustering. 15 Each implementation is optimized on CPU characteristics and vector units.…”

Section: Introductionmentioning

confidence: 99%

Performance evaluation of image convolution with WebAssembly

Oishi

Ishikawa

Nogami

et al. 2023

International Workshop on Advanced Imaging Technology (IWAIT) 2023

View full text Add to dashboard Cite

Writing image processing in WebAssembly enables the development of highly portable libraries. However, since WebAssembly is a new technology, there are not enough libraries available, and it is only supported by OpenCV. However, OpenCV has not been optimized specifically for WebAssembly. Therefore, in this paper, we develop native code using WebAssembly and compare it with OpenCV's Gaussian filter using the separable Gaussian filter, a classical acceleration method for Gaussian filters. Experimental results showed that the Separable Gaussian filter was faster than OpenCV by performing vector operations in WebAssembly.

show abstract

Tiling and PCA Strategy for Clustering-Based High-Dimensional Gaussian Filtering

Oishi,

Fukushima

2023

SN COMPUT. SCI.

View full text Add to dashboard Cite

Edge-preserving filtering is an essential tool for image processing applications and has various types of filtering. High-dimensional Gaussian filtering (HDGF) supports a wide range of edge-preserving filtering. This paper approximates HDGF by clustering with Nyström approximation, tiling, and principal component analysis (PCA) to accelerate HDGF. Also, we compare it with the conventional HDGF approximations and clarify its effective range. We accelerate HDGF by clustering-based constant-time algorithm, which has O(K) order for convolution, where K is the number of clusters. First, we perform PCA for dimensionality reduction and then cluster signals with k-means++. HDGF is decomposed to Gaussian filtering by approximate eigenvalue decomposition of Nyström approximation using the clusters. The Gaussian filtering is performed in a constant-time algorithm. The process is further accelerated by the tiling strategy cooperating with PCA. In our experimental results, we compared three approximated HDGFs: clustering-based HDGF, permutohedral lattice, and Gaussian KD-tree. Also, we evaluated six types of high dimensional signals: RGB, RGB-IR, RGB-D, flash/no-flash, hyperspectral image, and non-local means. The proposed clustering-based HDGF was effective for low/middle-dimensional cases: RGB, RGB-IR, RGB-D, flash/no-flash, and hyperspectral images. Also, tiling with PCA strategy is effective for the conventional permutohedral lattice and Gaussian KD-tree. In the approximation of HDGF, the clustering-based HDGF is the better solution for low/middle-dimensional signals. For the higher-dimensional case of non-local means filtering, the conventional HDGF of the permutohedral lattice with the proposed PCA tiling is effective. Our code and dataset are available at https://norishigefukushima.github.io/TilingPCA4CHDGF/.

show abstract

Vectorized implementation of K-means

Cited by 4 publications

References 8 publications

Performance evaluation of halide auto-scheduler with directional cubic convolution interpolation

Performance evaluation of halide auto-scheduler with directional cubic convolution interpolation

Performance evaluation of image convolution with WebAssembly

Tiling and PCA Strategy for Clustering-Based High-Dimensional Gaussian Filtering

Contact Info

Product

Resources

About