Realistic woven fabric rendering plays an important role in film production, video games, etc. The microflake model has recently been introduced for lightweight fabric rendering, which serves as the specular part of the woven fabric appearance. However, it cannot be used directly for real-time rendering due to the low sample count requirements for realtime rendering. The main challenge is a multi-scale representation which allows for efficient range query during rendering. To this end, we propose a multiscale representation of the specular lobes of each yarn segment using SGGX fitting. More specifically, we precompute the normal distribution functions (NDFs) of the yarn segments with different query sizes and then use several SGGX functions to represent the NDFs. During rendering, we aggregate the contribution for each pixel by querying the fitted SGGXs of yarn segments covered by the pixel's footprint. We also propose a bounding box representation for the yarn segments, enabling a practical intersection with the pixel's footprint. As a result, our method is able to render several typical types of woven fabrics in only 1–2 ms at 1080p resolution using an RTX 3090 video card. Our method outperforms existing approaches and shows closer rendering results to reference results.

Caustics rendering remains a long-standing challenge in Monte Carlo rendering because high-energy specular paths occupy only a small region of path space, making them difficult to sample effectively. Recent work such as Specular Manifold Sampling (SMS) [Zeltner et al. 2020] can stochastically sample these specular paths and estimate their unbiased weights using Bernoulli trials. However, applying SMS in interactive rendering is non-trivial because it is slow and delivers noisy images given a very limited time budget.

The DEM differencing method has been widely used for quantifying land surface change, especially in the field of glacier mass change. However, when the global DEM datasets are used to estimate large-scale glacier changes, the superposition effect of multi-source errors makes it difficult to remove the errors in the differencing maps using the existing method based on manually supervised correction. In this letter, to solve this issue, we propose a new stepwise correction method with an adaptive data partitioning strategy. The first step is to perform global spatial detrending, followed by global coarse co-registration. Subsequently, an initial data partitioning scheme based on the quadtree partitioning technique and relying on terrain slope is proposed. Furthermore, a parameter, i.e., the standardized standard deviation based on elevation changes, is used to further update the partitioning scheme. Finally, for each segmented data block, local spatial detrending is performed, followed by precise local co-registration. The proposed method was tested in the Pamir Plateau and the Tianshan River valley, based on the SRTM C-band and X-band DEMs and the Copernicus DEM. The final results show that the various types of error can be automatically and effectively removed, and that the normalized median absolute deviation and long-range correlation variance in ice-free areas shows an improvement of 10% and 24%, and 37% and 60% for the two study regions, respectively, demonstrating the good performance of the proposed method.

An advanced InSAR-based iterative tropospheric decomposition model for more effectively mitigating both vertically stratified and turbulent delays is proposed in this paper. The novelty of the model lies in two aspects: 1) incorporating turbulence variance, which characterizes the statistical properties of atmospheric turbulence, into a stochastic model when iteratively removing stratified delay, and 2) taking the distance and turbulence variance as the weight factors of samples, simultaneously, in spatially interpolating turbulence. In addition, our model is implemented interferogram by interferogram, so we do not need time-series InSAR datasets. The effectiveness of the proposed method is tested and validated using a set of simulated experiments and an in-site experiment over Hawaii Island, and the experimental results show that our method can mitigate tropospheric delays efficiently and robustly.

We propose an atmospheric turbulence statistical characteristic-enhanced stacking-interferometry synthetic aperture radar (InSAR) method to more accurately derive the average deformation velocity for slow or linear surface displacements. It can be used to estimate the variance and covariance matrix (VCM) for time-series InSAR observations pixel by pixel. Moreover, a minimum variance linear estimator is used to calculate optimal weights for interferogram stacking. Compared with traditional stacking, our method is an improvement, and its advantages include the following: 1) the relevance between time-series atmospheric delays of interferograms is considered; 2) the atmospheric turbulence is better suppressed; and 3) the performance is not limited by the quantity of interferograms. The effectiveness of the new method is verified through a series of simulated experiments and real Sentinel-1 experiments over Southern California. The results demonstrate that our proposed method is more effective and robust in removing turbulent atmospheres.

Due to the heterogeneity of workloads and the randomness and complexity of hybrid scheduling environment, How to address the challenge of minimizing data center running cost while ensuring workloads SLA has emerged as a significant research problem. To address this, we propose an Optimal Deep Reinforcement Learning Model for Running Cost Optimization in Hybrid Environments (ODRL). Firstly, we propose Running Cost Model (RCM) that analyzes data center running costs from two perspectives based on different running patterns of the workload and node affinity constraints, minimizing computing node utilization and resource running costs. Subsequently, we propose a Priority-Aware Scheduling Algorithm based on Deep Reinforcement Learning(PASD) that use AHP-TOPSIS to quantify workload characterization, get real-time feedback for environment and update historical experience through prioritized experience replay. Finally, extensive experimental results validate the effectiveness of the proposed model. In comparison to Kube-Scheduler, ODRL achieves a significant 15% improvement in reducing data center running cost. Additionally, there is a 7.8% increase in CPU utilization and a 16.7% increase in memory utilization.