Full-reference image quality metrics (FR-IQMs) aim to measure the visual differences between a pair of reference and distorted images, with the goal of accurately predicting human judgments. However, existing FR-IQMs, including traditional ones like PSNR and SSIM and even perceptual ones such as HDR-VDP, LPIPS, and DISTS, still fall short in capturing the complexities and nuances of human perception. In this work, rather than devising a novel IQM model, we seek to improve upon the perceptual quality of existing FR-IQM methods. We achieve this by considering visual masking, an important characteristic of the human visual system that changes its sensitivity to distortions as a function of local image content. Specifically, for a given FR-IQM metric, we propose to predict a visual masking model that modulates reference and distorted images in a way that penalizes the visual errors based on their visibility. Since the ground truth visual masks are difficult to obtain, we demonstrate how they can be derived in a self-supervised manner solely based on mean opinion scores (MOS) collected from an FR-IQM dataset. Our approach results in enhanced FR-IQM metrics that are more in line with human prediction both visually and quantitatively.
相關內容
Fully-strict fork-join parallelism is a powerful model for shared-memory programming due to its optimal time scaling and strong bounds on memory scaling. The latter is rarely achieved due to the difficulty of implementing continuation stealing in traditional High Performance Computing (HPC) languages -- where it is often impossible without modifying the compiler or resorting to non-portable techniques. We demonstrate how stackless coroutines (a new feature in C++20) can enable fully-portable continuation stealing and present libfork a lock-free fine-grained parallelism library, combining coroutines with user-space, geometric segmented-stacks. We show our approach is able to achieve optimal time/memory scaling, both theoretically and empirically, across a variety of benchmarks. Compared to openMP (libomp), libfork is on average 7.2x faster and consumes 10x less memory. Similarly, compared to Intel's TBB, libfork is on average 2.7x faster and consumes 6.2x less memory. Additionally, we introduce non-uniform memory access (NUMA) optimizations for schedulers that demonstrate performance matching busy-waiting schedulers.
This study investigates the applicability of Kirchhoff migration (KM) for a fast identification of unknown objects in a real-world limited-aperture inverse scattering problem. To demonstrate the theoretical basis for the applicability including unique determination of objects, the imaging function of the KM was formulated using a uniformly convergent infinite series of Bessel functions of integer order of the first kind based on the integral equation formula for the scattered field. Numerical simulations performed using the experimental Fresnel dataset are exhibited to achieve the theoretical results.
Bayesian statistical graphical models are typically either continuous and parametric (Gaussian, parameterized by the graph-dependent precision matrix with Wishart-type priors) or discrete and non-parametric (with graph-dependent structure of probabilities of cells and Dirichlet-type priors). We propose to break this dichotomy by introducing two discrete parametric graphical models on finite decomposable graphs: the graph negative multinomial and the graph multinomial distributions. These models interpolate between the product of univariate negative binomial laws and the negative multinomial distribution, and between the product of binomial laws and the multinomial distribution, respectively. We derive their Markov decomposition and present related probabilistic models representations. We also introduce graphical versions of the Dirichlet distribution and inverted Dirichlet distribution, which serve as conjugate priors for the two discrete graphical Markov models. We derive explicit normalizing constants for both graphical Dirichlet laws and demonstrate that their independence structure (a graphical version of neutrality) yields a strong hyper Markov property for both Bayesian models. We also provide characterization theorems for graphical Dirichlet laws via strong hyper Markov property. Finally, we develop a model selection procedure for the Bayesian graphical negative multinomial model with respective Dirichlet-type priors.
Evaluating the Expected Information Gain (EIG) is a critical task in many areas of computational science and statistics, necessitating the approximation of nested integrals. Available techniques for this problem based on Quasi-Monte Carlo (QMC) methods have primarily focused on enhancing the efficiency of the inner integral approximation. In this work, we introduce a novel approach that extends the scope of these efforts to address inner and outer expectations simultaneously. Leveraging the principles of Owen's scrambling, we develop a randomized quasi-Monte Carlo (RQMC) method that improves the approximation of nested integrals. We also indicate how to combine this methodology with Importance Sampling to address a measure concentration arising in the inner integral. Our RQMC method capitalizes on the unique structure of nested expectations to offer a more efficient approximation mechanism. By incorporating Owen's scrambling techniques, we handle integrands exhibiting infinite variation in the Hardy-Krause (HK) sense, paving the way for theoretically sound error estimates. We derive asymptotic error bounds for the bias and variance of our estimator. In addition, we provide nearly optimal sample sizes for the inner and outer RQMC approximations, which are helpful for the actual numerical implementations. We verify the quality of our estimator through numerical experiments in the context of Bayesian optimal experimental design. Specifically, we compare the computational efficiency of our RQMC method against standard nested Monte Carlo integration across two case studies: one in thermo-mechanics and the other in pharmacokinetics. These examples highlight our approach's computational savings and enhanced applicability, showcasing the advantages of estimating the Expected Information Gain with greater efficiency and reduced computational cost.
With advances in scientific computing and mathematical modeling, complex scientific phenomena such as galaxy formations and rocket propulsion can now be reliably simulated. Such simulations can however be very time-intensive, requiring millions of CPU hours to perform. One solution is multi-fidelity emulation, which uses data of different fidelities to train an efficient predictive model which emulates the expensive simulator. For complex scientific problems and with careful elicitation from scientists, such multi-fidelity data may often be linked by a directed acyclic graph (DAG) representing its scientific model dependencies. We thus propose a new Graphical Multi-fidelity Gaussian Process (GMGP) model, which embeds this DAG structure (capturing scientific dependencies) within a Gaussian process framework. We show that the GMGP has desirable modeling traits via two Markov properties, and admits a scalable algorithm for recursive computation of the posterior mean and variance along at each depth level of the DAG. We also present a novel experimental design methodology over the DAG given an experimental budget, and propose a nonlinear extension of the GMGP via deep Gaussian processes. The advantages of the GMGP are then demonstrated via a suite of numerical experiments and an application to emulation of heavy-ion collisions, which can be used to study the conditions of matter in the Universe shortly after the Big Bang. The proposed model has broader uses in data fusion applications with graphical structure, which we further discuss.
Fast and efficient simulations of metal additive manufacturing (AM) processes are highly relevant to exploring the full potential of this promising manufacturing technique. The microstructure composition plays an important role in characterizing the part quality and deriving mechanical properties. When complete parts are simulated, one often needs to resort to strong simplifications such as layer-wise heating due to the large number of simulated time steps compared to the small time step sizes. This article proposes a scan-resolved approach to the coupled thermo-microstructural problem. Building on a highly efficient thermal model, we discuss the implementation of a phenomenological microstructure model for the evolution of the three main constituents of Ti-6Al-4V: stable $\alpha_s$-phase, martensite $\alpha_m$-phase and $\beta$-phase. The implementation is tailored to modern hardware features using vectorization and fast approximations of transcendental functions. A performance model and numerical examples verify the high degree of optimization. We demonstrate the applicability and predictive power of the approach and the influence of scan strategy and geometry. Depending on the specific example, results can be obtained with moderate computational resources in a few hours to days. The numerical examples include a prediction of the microstructure on the full NIST AM Benchmark cantilever specimen.
Modality discrepancies have perpetually posed significant challenges within the realm of Automated Audio Captioning (AAC) and across all multi-modal domains. Facilitating models in comprehending text information plays a pivotal role in establishing a seamless connection between the two modalities of text and audio. While recent research has focused on closing the gap between these two modalities through contrastive learning, it is challenging to bridge the difference between both modalities using only simple contrastive loss. This paper introduces Enhance Depth of Text Comprehension (EDTC), which enhances the model's understanding of text information from three different perspectives. First, we propose a novel fusion module, FUSER, which aims to extract shared semantic information from different audio features through feature fusion. We then introduced TRANSLATOR, a novel alignment module designed to align audio features and text features along the tensor level. Finally, the weights are updated by adding momentum to the twin structure so that the model can learn information about both modalities at the same time. The resulting method achieves state-of-the-art performance on AudioCaps datasets and demonstrates results comparable to the state-of-the-art on Clotho datasets.
Merging satellite and gauge data with machine learning produces high-resolution precipitation datasets, but uncertainty estimates are often missing. We address this gap by benchmarking six algorithms, mostly novel for this task, for quantifying predictive uncertainty in spatial interpolation. On 15 years of monthly data over the contiguous United States (CONUS), we compared quantile regression (QR), quantile regression forests (QRF), generalized random forests (GRF), gradient boosting machines (GBM), light gradient boosting machines (LightGBM), and quantile regression neural networks (QRNN). Their ability to issue predictive precipitation quantiles at nine quantile levels (0.025, 0.050, 0.100, 0.250, 0.500, 0.750, 0.900, 0.950, 0.975), approximating the full probability distribution, was evaluated using quantile scoring functions and the quantile scoring rule. Feature importance analysis revealed satellite precipitation (PERSIANN (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks) and IMERG (Integrated Multi-satellitE Retrievals) datasets) as the most informative predictor, followed by gauge elevation and distance to satellite grid points. Compared to QR, LightGBM showed improved performance with respect to the quantile scoring rule by 11.10%, followed by QRF (7.96%), GRF (7.44%), GBM (4.64%) and QRNN (1.73%). Notably, LightGBM outperformed all random forest variants, the current standard in spatial interpolation with machine learning. To conclude, we propose a suite of machine learning algorithms for estimating uncertainty in interpolating spatial data, supported with a formal evaluation framework based on scoring functions and scoring rules.
There is an immediate need for creative ways to improve resource ef iciency given the dynamic nature of robust sensor networks and their increasing reliance on data-driven approaches.One key challenge faced is ef iciently managing large data files collected from sensor networks for example optimal beehive image and video data files. We of er a revolutionary paradigm that uses cutting-edge edge computing techniques to optimize data transmission and storage in order to meet this problem. Our approach encompasses data compression for images and videos, coupled with a data aggregation technique for numerical data. Specifically, we propose a novel compression algorithm that performs better than the traditional Bzip2, in terms of data compression ratio and throughput. We also designed as an addition a data aggregation algorithm that basically performs very well by reducing on the time to process the overhead of individual data packets there by reducing on the network traf ic. A key aspect of our approach is its ability to operate in resource-constrained environments, such as that typically found in a local beehive farm application from where we obtained various datasets. To achieve this, we carefully explore key parameters such as throughput, delay tolerance, compression rate, and data retransmission. This ensures that our approach can meet the unique requirements of robust network management while minimizing the impact on resources. Overall, our study presents and majorly focuses on a holistic solution for optimizing data transmission and processing across robust sensor networks for specifically local beehive image and video data files. Our approach has the potential to significantly improve the ef iciency and ef ectiveness of robust sensor network management, thereby supporting sustainable practices in various IoT applications such as in Bee Hive Data Management.
We present ResMLP, an architecture built entirely upon multi-layer perceptrons for image classification. It is a simple residual network that alternates (i) a linear layer in which image patches interact, independently and identically across channels, and (ii) a two-layer feed-forward network in which channels interact independently per patch. When trained with a modern training strategy using heavy data-augmentation and optionally distillation, it attains surprisingly good accuracy/complexity trade-offs on ImageNet. We will share our code based on the Timm library and pre-trained models.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191