GPU Devices Research Articles

Study on Large-Scale Urban Water Distribution Network Computation Method Based on a GPU Framework

Large-scale urban water distribution network simulation plays a critical role in the construction, monitoring, and maintenance of urban water distribution systems. However, during the simulation process, matrix inversion calculations generate a large amount of computational data and consume significant amounts of time, posing challenges for practical applications. To address this issue, this paper proposes a parallel gradient calculation algorithm based on GPU hardware and the CUDA Toolkit library and compares it with the EPANET model and a model based on CPU hardware and the Armadillo library. The results show that the GPU-based model not only achieves a precision level very close to the EPANET model, reaching 99% accuracy, but also significantly outperforms the CPU-based model. Furthermore, during the simulation, the GPU architecture is able to efficiently handle large-scale data and achieve faster convergence, significantly reducing the overall simulation time. Particularly in handling larger-scale water distribution networks, the GPU architecture can improve computational efficiency by up to 13 times. Further analysis reveals that different GPU models exhibit significant differences in computational efficiency, with memory capacity being a key factor affecting performance. GPU devices with larger memory capacity demonstrate higher computational efficiency when processing large-scale water distribution networks. This study demonstrates the advantages of GPU acceleration technology in the simulation of large-scale urban water distribution networks and provides important theoretical and technical support for practical applications in this field. By carefully selecting and configuring GPU devices, the computational efficiency of large-scale water distribution networks can be significantly improved, providing more efficient solutions for future urban water resource management and planning.

TLPGNN: A Lightweight Two-level Parallelism Paradigm for Graph Neural Network Computation on Single and Multiple GPUs

Graph Neural Networks (GNNs) are an emerging class of deep learning models specifically designed for graph-structured data. They have been effectively employed in a variety of real-world applications, including recommendation systems, drug development, and analysis of social networks. The GNN computation includes regular neural network operations and general graph convolution operations, which take most of the total computation time. Though several recent works have been proposed to accelerate the computation for GNNs, they face the limitations of heavy pre-processing, low efficiency atomic operations, and unnecessary kernel launches. In this article, we designTLPGNN, a lightweight two-level parallelism paradigm for GNN computation. First, we conduct a systematic analysis of the hardware resource usage of GNN workloads to understand the characteristics of GNN workloads deeply. With the insightful observations, we then divide the GNN computation into two levels, i.e.,vertex parallelismfor the first level andfeature parallelismfor the second. Next, we employ a novel hybrid dynamic workload assignment to address the imbalanced workload distribution. Furthermore, we fuse the kernels to reduce the number of kernel launches and cache the frequently accessed data into registers to avoid unnecessary memory traffic. To scaleTLPGNNto multi-GPU environments, we propose an edge-aware row-wise 1-D partition method to ensure a balanced workload distribution across different GPU devices. Experimental results on various benchmark datasets demonstrate the superiority of our approach, achieving substantial performance improvement over state-of-the-art GNN computation systems, including Deep Graph Library (DGL), GNNAdvisor, and FeatGraph, with speedups of 6.1×, 7.7×, and 3.0×, respectively, on average. Evaluations of multiple-GPUTLPGNNalso demonstrate that our solution achieves both linear scalability and a well-balanced workload distribution.

Nuance: Efficient Detection of Planets Transiting Active Stars

The detection of planetary transits in the light curves of active stars, featuring correlated noise in the form of stellar variability, remains a challenge. Depending on the noise characteristics, we show that the traditional technique that consists of detrending a light curve before searching for transits alters their signal-to-noise ratio and hinders our capability to discover exoplanets transiting rapidly rotating active stars. We present nuance, an algorithm to search for transits in light curves while simultaneously accounting for the presence of correlated noise, such as stellar variability and instrumental signals. We assess the performance of nuance on simulated light curves as well as on the Transiting Exoplanet Survey Satellite light curves of 438 rapidly rotating M dwarfs. For each data set, we compare our method to five commonly used detrending techniques followed by a search with the Box-Least-Squares algorithm. Overall, we demonstrate that nuance is the most performant method in 93% of cases, leading to both the highest number of true positives and the lowest number of false-positive detections. Although simultaneously searching for transits while modeling correlated noise is expected to be computationally expensive, we make our algorithm tractable and available as the JAX-powered Python package nuance, allowing its use on distributed environments and GPU devices. Finally, we explore the prospects offered by the nuance formalism and its use to advance our knowledge of planetary systems around active stars, both using space-based surveys and sparse ground-based observations.

Characterizing Parameter Scaling with Quantization for Deployment of CNNs on Real-Time Systems

Modern deep-learning models tend to include billions of parameters, reducing real-time performance. Embedded systems are compute-constrained while frequently used to deploy these models for real-time systems given size, weight, and power requirements. Tools like parameter-scaling methods help to shrink models to ease deployment. This research compares two scaling methods for convolutional neural networks, uniform scaling and NeuralScale, and analyzes their impact on inference latency, memory utilization, and power. Uniform scaling scales the number of filters evenly across a network. NeuralScale adaptively scales the model to theoretically achieve the highest accuracy for a target parameter count. In this study, VGG-11, MobileNetV2, and ResNet-50 models were scaled to four ratios: 0.25×, 0.50×, 0.75×, 1.00×. These models were benchmarked on an ARM Cortex-A72 CPU, an NVIDIA Jetson AGX Xavier GPU, and a Xilinx ZCU104 FPGA. Additionally, quantization was applied to meet real-time objectives. The CIFAR-10 and tinyImageNet datasets were studied. On CIFAR-10, NeuralScale creates more computationally intensive models than uniform scaling for the same parameter count, with relative speeds of 41% on the CPU, 72% on the GPU, and 96% on the FPGA. The additional computational complexity is a tradeoff for accuracy improvements in VGG-11 and MobileNetV2 NeuralScale models but reduced ResNet-50 NeuralScale accuracy. Furthermore, quantization alone achieves similar or better performance on the CPU and GPU devices when compared with models scaled to 0.50×, despite slight reductions in accuracy. On the GPU, quantization reduces latency by 2.7× and memory consumption by 4.3×. Uniform-scaling models are 1.8× faster and use 2.8× less memory. NeuralScale reduces latency by 1.3× and dropped memory by 1.1×. We find quantization to be a practical first tool for improved performance. Uniform scaling can easily be applied for additional improvements. NeuralScale may improve accuracy but tends to negatively impact performance, so more care must be taken with it.

Research on Bearing Surface Scratch Detection Based on Improved YOLOV5.

Bearings are crucial components of machinery and equipment, and it is essential to inspect them thoroughly to ensure a high pass rate. Currently, bearing scratch detection is primarily carried out manually, which cannot meet industrial demands. This study presents research on the detection of bearing surface scratches. An improved YOLOV5 network, named YOLOV5-CDG, is proposed for detecting bearing surface defects using scratch images as targets. The YOLOV5-CDG model is based on the YOLOV5 network model with the addition of a Coordinate Attention (CA) mechanism module, fusion of Deformable Convolutional Networks (DCNs), and a combination with the GhostNet lightweight network. To achieve bearing surface scratch detection, a machine vision-based bearing surface scratch sensor system is established, and a self-made bearing surface scratch dataset is produced as the basis. The scratch detection final Average Precision (AP) value is 97%, which is 3.4% higher than that of YOLOV5. Additionally, the model has an accuracy of 99.46% for detecting defective and qualified products. The average detection time per image is 263.4 ms on the CPU device and 12.2 ms on the GPU device, demonstrating excellent performance in terms of both speed and accuracy. Furthermore, this study analyzes and compares the detection results of various models, demonstrating that the proposed method satisfies the requirements for detecting scratches on bearing surfaces in industrial settings.

A Convolutional Neural Network for Beamforming and Image Reconstruction in Passive Cavitation Imaging.

Convolutional neural networks (CNNs), initially developed for image processing applications, have recently received significant attention within the field of medical ultrasound imaging. In this study, passive cavitation imaging/mapping (PCI/PAM), which is used to map cavitation sources based on the correlation of signals across an array of receivers, is evaluated. Traditional reconstruction techniques in PCI, such as delay-and-sum, yield high spatial resolution at the cost of a substantial computational time. This results from the resource-intensive process of determining sensor weights for individual pixels in these methodologies. Consequently, the use of conventional algorithms for image reconstruction does not meet the speed requirements that are essential for real-time monitoring. Here, we show that a three-dimensional (3D) convolutional network can learn the image reconstruction algorithm for a 16×16 element matrix probe with a receive frequency ranging from 256 kHz up to 1.0 MHz. The network was trained and evaluated using simulated data representing point sources, resulting in the successful reconstruction of volumetric images with high sensitivity, especially for single isolated sources (100% in the test set). As the number of simultaneous sources increased, the network's ability to detect weaker intensity sources diminished, although it always correctly identified the main lobe. Notably, however, network inference was remarkably fast, completing the task in approximately 178 s for a dataset comprising 650 frames of 413 volume images with signal duration of 20μs. This processing speed is roughly thirty times faster than a parallelized implementation of the traditional time exposure acoustics algorithm on the same GPU device. This would open a new door for PCI application in the real-time monitoring of ultrasound ablation.

Interference-aware opportunistic job placement for shared distributed deep learning clusters

Distributed deep learning frameworks facilitate large deep learning workloads. These frameworks support sharing one GPU device among multiple jobs to improve resource utilization. Modern deep learning training jobs consume a large amount of GPU memory. Despite that, sharing GPU memory among jobs is still possible because a training job has iterative steps that its memory usage fluctuates over time. However, resource sharing also introduces the risk of job performance degradation. Co-located jobs sharing a GPU device may suffer from different levels of interference, mainly caused by memory oversharing. How to improve resource utilization while maintaining good job performance is a novel challenge for job placement strategies. This paper studies the job placement problem. We propose an opportunistic memory sharing model to describe the time-varying job memory requirements. Based on this model, we introduce an Opportunistic Job Placement Problem (OJPP) for shared GPU clusters that seek job placement configurations using a minimum number of GPU devices and guarantee user-defined performance requirements at the same time. We propose a greedy algorithm and a heuristic algorithm with computational complexities of O(nlog⁡n) and O(n2log⁡n), respectively, to solve the problem. We also propose an online adjustment algorithm with the computational complexity of O(nlog⁡n) to perform updates to job placement configurations in runtime. A machine-learning-based interference prediction method is used to prepare accurate interference estimations. Extensive experiments are conducted on a GPU cluster to verify the correctness and effectiveness of our algorithms. Compared with standalone training jobs on dedicated clusters, the proposed approach reduces resource consumption by 46% in a shared cluster, while guaranteeing over 92.97% of the job performance, in terms of average job completion time.

Dragon: A multi-GPU orbital-free density functional theory molecular dynamics simulation package for modeling of warm dense matter

As progress in electronic structure theoretical methods is made, ab initio molecular dynamics (MD) based on orbital-free density functional theory (OF-DFT) is becoming increasingly more successful at substituting the traditional, very accurate but computationally costly Kohn-Sham (KS) approach for simulations of matter at the challenging warm dense matter (WDM) regime. However, despite the significant cost alleviation of eliminating the dependence on the KS orbitals, OF-DFT MD runs require ∼102 to 103 CPU cores running for days, or even weeks, for simulations of systems comprised of 102 to 103 atoms, depending on thermodynamic conditions. We present Dragon, a multi-GPU OF-DFT MD code for fast and efficient simulations of WDM. With a relatively small allocation of resources (4 to 8 GPU devices) it can provide an order of magnitude speedup for simulations containing O(104) atoms and target systems composed of O(105) atoms at conditions within the WDM regime, which is currently outside the capabilities of CPU codes.

Multi-GPUs DEM algorithm and its application in the simulation of granular materials

The Discrete Element Method (DEM) is one of the most common numerical approach for modeling the physical and mechanical properties of granular materials. However, its computational efficiency can be a major bottleneck, which significantly hinders its application in engineering fields. To overcome this challenge, a multi-GPU framework with OpenMP has been developed. In this framework, the DEM particles in 3D space and their corresponding contact information are sorted in a given direction in the simulation, and each GPU device handles a subdomain of particles and relative contacts. Additionally, the contact information is densely arranged to minimize memory usage. To further enhance efficiency, sorting and packing algorithms for the contact information have been introduced. During calculating loop, the number of contacts on each device is monitored, and the spatial division of devices is adjusted to balance memory distribution. The efficacy of the multi-GPU algorithm is verified through a chute flow test. Using numerical results, the segregation process and mechanics of granular materials during sliding have been studied in detail. Furthermore, the simulations demonstrate that the developed multi-GPU algorithm greatly enhances computational efficiency and reduces memory usage per device.

Neural networks for fatigue crack propagation predictions in real-time under uncertainty

Crack propagation analyses are fundamental for all mechanical structures for which safety must be guaranteed, e.g. as for the aviation and aerospace fields. The estimation of life for structures in presence of defects is a process inevitably affected by numerous and unavoidable uncertainty and variability sources, whose effects need to be quantified to avoid unexpected failures or excessive conservativism.In this work, residual fatigue life prediction models have been created through neural networks for the purpose of performing probabilistic life predictions of damaged structures in real-time and under stochastically varying input parameters. In detail, five different neural network architectures have been compared in terms of accuracy, computational runtimes and minimum number of samples needed for training, so to determine the ideal architecture with the strongest generalization power. The networks have been trained, validated and tested by using the fatigue life predictions computed by means of simulations developed with FEM and Monte Carlo methods. A real-world case study has been presented to show how the proposed approach can deliver accurate life predictions even when input data are uncertain and highly variable.Results demonstrated that the “H1-L1” neural network has been the best model, achieving an accuracy (Mean Square Error) of 4.8e-7 on the test dataset, and the best and the most stable results when decreasing the amount of data. Additionally, since requiring only very few parameters, its potential applicability for Structural Health Monitoring purposes in small cost-effective GPU devices resulted to be attractive.

PipePar: Enabling fast DNN pipeline parallel training in heterogeneous GPU clusters

Recently, pipeline parallelism for large-scale Deep Neural Network (DNN) training has been developed, which partitions the DNN model across multiple devices (e.g., GPUs) and improves the training efficiency by processing data divided into minibatches as a pipeline. However, existing model partitioning algorithms are mostly designed for homogeneous clusters with the same GPU devices and network connections (e.g., bandwidths), while heterogeneous GPU clusters are widely used in mainstream computing infrastructures. In heterogeneous environment, devices are equipped with different GPUs and network connections, and the efficiency of previous approaches is unsatisfactory due to the unbalanced load of the pipeline stages. In this paper, we propose PipePar, a model partitioning and task placement algorithm for pipeline parallel DNN training in heterogeneous GPU clusters. PipePar is based on dynamic programming with search space pruning that takes into consideration both the heterogeneity of GPUs and network bandwidth. PipePar can profile the DNN model for each type of GPU and conduct model partitioning and task placement based on given GPUs and network connections, which can optimize pipeline load balancing in heterogeneous environments and thus improve training efficiency. We design and implement a pipeline-based distributed deep learning training system in a heterogeneous GPU cluster and show through extensive experiments that PipePar outperforms the baseline approaches in the speed of large-scale DNN training.

On-Board Deep-Learning-Based Unmanned Aerial Vehicle Fault Cause Detection and Classification via FPGAs

With the increase in the use of unmanned aerial vehicles (UAVs)/drones, it is important to detect and identify causes of failure in real time for proper recovery from a potential crash-like scenario or postincident forensics analysis. The cause of crash could be either a fault in the sensor/actuator system, a physical damage/attack, or a cyber attack on the drone's software. In this work, we propose novel architectures based on deep convolutional and long short-term memory neural networks to detect (via autoencoder) and classify drone misoperations based on real-time sensor data. The proposed architectures are able to learn high-level features automatically from the raw sensor data. Empirical results show that our solution is able to detect (with over 90% accuracy) and classify various types of drone misoperations [with about 99% accuracy (simulation data) and up to 85% accuracy (experimental data)]. Furthermore, with the help of field programmable gate array-based hardware acceleration, we achieved a speedup of 40x ( <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\sim\!\text{2.6 ms}$</tex-math></inline-formula> ) for detection, while consuming half the amount of power compared with onboard GPU devices, such as NVIDIA Jetson TX2.

GPU-Based Algorithms for Processing the k Nearest-Neighbor Query on Spatial Data Using Partitioning and Concurrent Kernel Execution

Algorithms for answering the k nearest-neighbor (k-NN) query are widely used for queries in spatial databases and for distance classification of a group of query points against a reference dataset to derive the dominating feature class. GPU devices have significantly more processing cores than CPUs and faster device memory than the main memory accessed by CPUs, thus, providing higher computing power for processing demanding queries like the k-NN. However, since device and/or main memory may not be able to host an entire, rather big, reference and query datasets, storing these datasets in a fast secondary device, like a solid state disk (SSD), and partially retrieve the required, at each stage, partitions is, in many practical cases, a feasible solution. We propose and implement the first GPU-based algorithms for processing the k-NN query for big reference and query spatial data stored on SSDs. Based on 3d synthetic and real big spatial data, we experimentally compare these algorithms and highlight the most efficient algorithmic variation. This variation utilizes a CUDA feature known as Concurrent Kernel Execution, to further improve its performance.

Quantization-Aware Interval Bound Propagation for Training Certifiably Robust Quantized Neural Networks

We study the problem of training and certifying adversarially robust quantized neural networks (QNNs). Quantization is a technique for making neural networks more efficient by running them using low-bit integer arithmetic and is therefore commonly adopted in industry. Recent work has shown that floating-point neural networks that have been verified to be robust can become vulnerable to adversarial attacks after quantization, and certification of the quantized representation is necessary to guarantee robustness. In this work, we present quantization-aware interval bound propagation (QA-IBP), a novel method for training robust QNNs. Inspired by advances in robust learning of non-quantized networks, our training algorithm computes the gradient of an abstract representation of the actual network. Unlike existing approaches, our method can handle the discrete semantics of QNNs. Based on QA-IBP, we also develop a complete verification procedure for verifying the adversarial robustness of QNNs, which is guaranteed to terminate and produce a correct answer. Compared to existing approaches, the key advantage of our verification procedure is that it runs entirely on GPU or other accelerator devices. We demonstrate experimentally that our approach significantly outperforms existing methods and establish the new state-of-the-art for training and certifying the robustness of QNNs.

TransVOD: End-to-End Video Object Detection With Spatial-Temporal Transformers.

Detection Transformer (DETR) and Deformable DETR have been proposed to eliminate the need for many hand-designed components in object detection while demonstrating good performance as previous complex hand-crafted detectors. However, their performance on Video Object Detection (VOD) has not been well explored. In this paper, we present TransVOD, the first end-to-end video object detection system based on simple yet effective spatial-temporal Transformer architectures. The first goal of this paper is to streamline the pipeline of current VOD, effectively removing the need for many hand-crafted components for feature aggregation, e.g., optical flow model, relation networks. Besides, benefited from the object query design in DETR, our method does not need post-processing methods such as Seq-NMS. In particular, we present a temporal Transformer to aggregate both the spatial object queries and the feature memories of each frame. Our temporal transformer consists of two components: Temporal Query Encoder (TQE) to fuse object queries, and Temporal Deformable Transformer Decoder (TDTD) to obtain current frame detection results. These designs boost the strong baseline deformable DETR by a significant margin (3 %-4 % mAP) on the ImageNet VID dataset. TransVOD yields comparable performances on the benchmark of ImageNet VID. Then, we present two improved versions of TransVOD including TransVOD++ and TransVOD Lite. The former fuses object-level information into object query via dynamic convolution while the latter models the entire video clips as the output to speed up the inference time. We give detailed analysis of all three models in the experiment part. In particular, our proposed TransVOD++ sets a new state-of-the-art record in terms of accuracy on ImageNet VID with 90.0 % mAP. Our proposed TransVOD Lite also achieves the best speed and accuracy trade-off with 83.7 % mAP while running at around 30 FPS on a single V100 GPU device. Code and models are available at https://github.com/SJTU-LuHe/TransVOD.

PaScaL_TDMA 2.0: A multi-GPU-based library for solving massive tridiagonal systems

PaScaL_TDMA 2.0: A multi-GPU-based library for solving massive tridiagonal systems

2D Numerical Simulation of Floods in Ebro River and Analysis of Boundary Conditions to Model the Mequinenza Reservoir Dam

The computational simulation of rivers is a useful tool that can be applied in a wide range of situations from providing real time alerts to the design of future mitigation plans. However, for all the applications, there are two important requirements when modeling river behavior: accuracy and reasonable computational times. This target has led to recent developments in numerical models based on the full two-dimensional (2D) shallow water equations (SWE). This work presents a GPU accelerated 2D SW model for the simulation of flood events in real time. It is based on a well-balanced explicit first-order finite volume scheme able to run over dry beds without the numerical instabilities that are likely to occur when used in complex topography. The model is applied to reproduce a real event in the reach of the Ebro River (Spain) with a downstream reservoir, in which a study of the most appropriate boundary condition (BC) for modeling of the dam is assessed (time-dependent level condition and weir condition). The whole creation of the model is detailed in terms of mesh optimization and validation. The simulation results are compared with field data over the flood duration (up to 20 days), allowing an analysis of the performance and time saved by different GPU devices and with the different BCs. The high values of fit between observed and simulated results, as well as the computational times achieved, are encouraging to propose the use of the model as a forecasting system.

Rice grains and grain impurity segmentation method based on a deep learning algorithm-NAM-EfficientNetv2

An appropriate image segmentation algorithm is required for discriminating between full grains and grain impurities. In this study, a lightweight fully convolutional segmentation algorithm based on NAM EfficientNetV2 was proposed to solve the problem of slow processing speed in mobile terminal equipment owing to limited computing resources and a large number of model parameters and improve the detection accuracy. First, a standardized NAM attention mechanism was introduced to replace the SE attention mechanism used in EfficientNetV2 and the improved NAM-EfficientNetV2 network was used as a feature extraction structure. Then, in the up-sampling process, the multi-scale features output by the shallow network are fused to effectively use low-level semantics to encode spatial details and fully convolutional pixel segmentation technology is used to achieve rice grain and impurity segmentation. Finally, compared with the baseline model, the detection accuracy of the model was further improved on a self-made dataset. The comprehensive evaluation index F1 of rice grain and its impurities were 95.26% and 93.27%, respectively, and the model parameter amount was 20.6 M. Combined with post-processing, detecting an image on the GPU device took an average of 0.103 s and 0.301 s on the CPU device. The experimental results showed that the improved algorithm is more lightweight, which provides a reference for the model to be deployed in mobile terminal equipment to realize the function of real-time detection of grain impurity in the combine harvester.

High-order discontinuous Galerkin hydrodynamics with sub-cell shock capturing on GPUs

ABSTRACT Hydrodynamical numerical methods that converge with high-order hold particular promise for astrophysical studies, as they can in principle reach prescribed accuracy goals with higher computational efficiency than standard second- or third-order approaches. Here we consider the performance and accuracy benefits of Discontinuous Galerkin (DG) methods, which offer a particularly straightforward approach to reach extremely high order. Also, their computational stencil maps well to modern GPU devices, further raising the attractiveness of this approach. However, a traditional weakness of this method lies in the treatment of physical discontinuities such as shocks. We address this by invoking an artificial viscosity field to supply required dissipation where needed, and which can be augmented, if desired, with physical viscosity and thermal conductivity, yielding a high-order treatment of the Navier–Stokes equations for compressible fluids. We show that our approach results in sub-cell shock capturing ability, unlike traditional limiting schemes that tend to defeat the benefits of going to high order in DG in problems featuring many shocks. We demonstrate exponential convergence of our solver as a function of order when applied to smooth flows, such as the Kelvin–Helmholtz reference problem of Lecoanet et al. We also demonstrate excellent scalability of our GPU implementation up to hundreds of GPUs distributed on different compute nodes. In a first application to driven, subsonic turbulence, we highlight the accuracy advantages of high-order DG compared to traditional second-order accurate methods, and we stress the importance of physical viscosity for obtaining accurate velocity power spectra.

Benchmark on the accuracy and efficiency of several neural network based phase pickers using datasets from China Seismic Network

Seismic phase pickers based on deep neural networks have been extensively used recently, demonstrating their advantages on both performance and efficiency. However, these pickers are trained with and applied to different data. A comprehensive benchmark based on a single dataset is therefore lacking. Here, using the recently released DiTing dataset, we analyzed performances of seven phase pickers with different network structures, the efficiencies are also evaluated using both CPU and GPU devices. Evaluations based on F1-scores reveal that the recurrent neural network (RNN) and EQTransformer exhibit the best performance, likely owing to their large receptive fields. Similar performances are observed among PhaseNet (UNet), UNet++, and the lightweight phase picking network (LPPN). However, the LPPN models are the most efficient. The RNN and EQTransformer have similar speeds, which are slower than those of the LPPN and PhaseNet. UNet++ requires the most computational effort among the pickers. As all of the pickers perform well after being trained with a large-scale dataset, users may choose the one suitable for their applications. For beginners, we provide a tutorial on training and validating the pickers using the DiTing dataset. We also provide two sets of models trained using datasets with both 50 Hz and 100 Hz sampling rates for direct application by end-users. All of our models are open-source and publicly accessible.