Segment anything model (SAM), an eminent universal image segmentation model, has recently gathered considerable attention within the domain of medical segmentation. Despite remarkable performance SAM on natural images, it grapples with significant degradation and limited generalization when confronted particularly those involving objects low contrast, faint boundaries, intricate shapes, diminutive sizes. In this paper, we propose SAMUS, a tailored for ultrasound contrast to previous...

Fiber bend is a major challenge of multimode fiber (MMF) imaging. More robustness against demonstrated in compressive MMF imaging using mean speckle patterns captured at multiple potential bending configurations beforehand, rather than sticking to single initial configuration. Experiments demonstrate an overall quality improvement on recovered images previous work, which important for robust endoscopic application.

In this paper, we present a post-editing scheme for camera-work. It is based on video retargeting, but aims to enhance the aesthetic interest of home produced sequences. The essential part clips are emphasized by automatically zooming in. camera pans preserve important features within frame while Different from traditional retargeting schemes, use variable factor which motion saliency frames.

In this paper, we present an algorithm that can successfully improve the quality of hazy images and offer visually-pleasant haze-free results with vivid colors. We propose full-saturation assumption (FSA) based on aesthetic photographic effect: photos colors are visually pleasant, first recover degraded saturation layer. The depth image is also obtained as a by-product. then apply example-based approach to avoid over-saturation. proposed serve criterion evaluate degree haze removal. Results...

Convolutional neural networks have proved their capability over wide areas of computer vision and has lead to superlative performance in the benchmarking tasks. Efforts train run deep is computationally expensive requires a cloud infrastructure equiped with appropriate hardware. In this paper, we introduce technical details results from our service for image analysis. It includes ConvNet region proposals, able cater requests range object classification detection Our approach also encompasses...

Truncated singular value decomposition (TSVD) is a popular recovery algorithm for multimode fiber single-pixel imaging (MMF-SPI), and it uses truncation thresholds to suppress noise influences. However, due the sensitivity of MMF relative stochastic disturbances, threshold requires frequent re-determination as levels dynamically fluctuate. In response, we design an adaptive determination (ATTD) method TSVD-based MMF-SPI in disturbed environments. Simulations experiments reveal that ATTD...

Rugby is a fair compete fullfed with human discipline, self-control, and mutual respect between athletes. Notably, tackle action one of the most basic rugby actions smallest unit body touch confrontation when attacker defender fight for possession rugby. After match, trainer usually count number happened during match to analyze discuss adjusting tactics. This paper studies method that optimized BP neural network based on genetic algorithm build classifier through extracting 10-dimension gray...

In this paper, we present a novel automatic image slideshow system that explores new medium between images and music. It can be regarded as selection composition criterion. Based on the idea of ``hearing colors, seeing sounds" from art music visualization, equal importance is assigned to features audio properties for better synchronization. We minimize aesthetic energy distance visual features. Given set images, subset selected by correlating with input properties. The are then synchronized...

Through the variation of optimal truncation for multimode fiber single-pixel imaging (MMF-SPI) using truncated singular value decomposition, source and level system noise could be determined estimated which is desired MMF-SPI.

Nonlinear interpolation is proposed to extend the scope of generalized Wigner-Smith operator looking for perturbation-invariant modes in multimode fiber, found those under unknown configuration from quadratic intensity transmission matrix.

Proposed SSIM-Edge-CNN-U 2 Net model improves loss function and adopts two-stage residual learning, achieving higher quality of restored natural scene images from speckles through multimode fiber.

Single pixel imaging experiment through single multimode fiber demonstrates capability of multi-millimeter depth field, almost three orders larger than previous work, also robustness against configuration change caused by bending. © 2019 The Author(s)

More robustness against fiber bend is demonstrated in compressive multimode imaging using mean speckle patterns captured at multiple potential bending configurations beforehand, rather than sticking to single initial configuration.

Abstract Rugby is a highly challenging and spectator contact team sport with intense physical confrontation, specific technique, group tactics, which requires athletes have tenacious will, courage teamwork. “Tackle” an essential feature of rugby the smallest action unit body confrontation between offensive defensive players during match. Tackling ability not only demonstration personal skills, courage, experience, but also base tacitly cooperating other teammates in helpful defense...

A real-valued neural network overwhelmingly outyerforms its complex-valued counterpart in natural scene reconstruction through multimode fiber, suggesting that knowledge of intensity rather than phase information is enough, therefore brings much robustness and convenience practice.

For multimode fiber, modified optimization goal stressing on transmission matrix diagonalization improves method self-consistency of searching propagation-invariant modes, assuming the disturbance being Zernike aberration.

Numerical simulation recommends coherently illuminating multiple cores, other than increasing core number, as the way to get enough projecting patterns for high resolution single pixel imaging endoscopy using multicore fiber and photonic lantern.

Utility of light intensity based linear approximation image transmission through multimode fiber is undermined by simple neural network modeling, i.e., single full connection layer with ReLU activation, necessitates inclusion nonlinearity.