Vector Quantization Research Articles

Research on control sequence of work mode switching for planetary row multi-mode hybrid powertrain system

According to the characteristics of the planetary row hybrid power system, it is an important issue for the system to adopt a reasonable working mode to realize the efficient operation of the hybrid power system in different operating conditions. Paper takes the single-row hybrid system as the research object. To address the complexity of the working model, a self-construction method using vehicle modeling is proposed. State-space equations are developed for each mode, and an objective function model integrating main and sub-mode switching rules is constructed using the DP algorithm. Learning Vector Quantization (LVQ) network clustering extracts mode-switching sequences. Urban and highway cycles (UDDS and HWFET) are chosen as global optimization parameters. The DP algorithm’s global optimization is applied, extracting sequence basis from UDDS and HWFET compound conditions. Utilizing LVQ input data, the main mode’s power demand is divided into 4 sub-mode regions, generating a switching sequence diagram. Joint simulations verify the effectiveness. ECMS control with regular mode switching yields 4.7458 L/100 km fuel consumption under UDDS & HWFET, 5.53% higher than DP optimization. WLTC conditions, with more medium and high-speed operation, consumption is 5.031 L/100 km, offering further optimization potential.

A Review of Facial Generation Technology Research

Abstract. Face generation, as a cutting-edge generative technology, has made significant strides in various image synthesis tasks, including facial inpainting, text-to-image translation, and video-based facial animation. Generating realistic and diverse human faces is a critical task in computer vision, with a wide array of real-world applications. Given the remarkable success of face generation models, there has been a growing interest in leveraging advanced techniques to further improve the quality, diversity, and control of generated faces. This paper will provide a comprehensive overview of the state-of-the-art face generation techniques. Specifically, the paper reviews the key approaches in this field, including Generative Adversarial Networks (GANs), Vector Quantized methods, and the more recent Diffusion Models, each of which has contributed significantly to the advancement of face generation. The paper discusses how these models have evolved to handle the complexity of face generation, from capturing subtle facial details to enabling fine-grained control over facial attributes. The paper also explores the major applications of face generation technologies, with a particular emphasis on their use in entertainment, virtual reality, and security. Finally, the paper identifies promising directions for future research in face generation, such as improving the interpretability of models, addressing ethical concerns, and enhancing the ability to generate faces that are both highly realistic and diverse.

Recognition of Impact Load on Connecting-Shaft Rotor System Based on Motor Current Signal Analysis.

Impact loads affect the operational performance and safety life of rolling equipment's connecting-shaft rotor system, even causing faults and accidents. Therefore, recognizing and investigating impact loads is of great significance. Hence, a load recognition method based on motor current information is proposed in this paper to recognize impact loads on the connecting-shaft rotor system. First, the fast Fourier transform is used to obtain the frequency domain information for the motor's current response signal from the rotor system load recognition test. Consequently, the required load response information can be presented more clearly using the singular value decomposition method to remove the power frequency components in the current signal. Then, wavelet packet decomposition is performed on the signal to generate energy analysis feature vectors. A qualitative recognition of the impact load on the system is achieved by learning vector quantization neural networks; the resulting load recognition results are good. These findings indicate that using the motor current as the analysis signal can solve the problem of the difficult layout for traditional vibration sensors in rolling sites. The preprocessing and recognition method of the current response signal can recognize the impact load, confirming the applicability and feasibility of the proposed method.

Lossless Recompression of Vector Quantization Index Table for Texture Images Based on Adaptive Huffman Coding Through Multi-Type Processing

With the development of the information age, all walks of life are inseparable from the internet. Every day, huge amounts of data are transmitted and stored on the internet. Therefore, to improve transmission efficiency and reduce storage occupancy, compression technology is becoming increasingly important. Based on different application scenarios, it is divided into lossless data compression and lossy data compression, which allows a certain degree of compression. Vector quantization (VQ) is a widely used lossy compression technology. Building upon VQ compression technology, we propose a lossless compression scheme for the VQ index table. In other words, our work aims to recompress VQ compression technology and restore it to the VQ compression carrier without loss. It is worth noting that our method specifically targets texture images. By leveraging the spatial symmetry inherent in these images, our approach generates high-frequency symbols through difference calculations, which facilitates the use of adaptive Huffman coding for efficient compression. Experimental results show that our scheme has better compression performance than other schemes.

A selective LVQ algorithm for improving instance reduction techniques and its application for text classification

Instance-Based Learning, such as the k Nearest Neighbor (kNN), offers a straightforward and effective solution for text classification. However, as a lazy learner, kNN’s performance heavily relies on the quality and quantity of training instances, often leading to time and space inefficiencies. This challenge has spurred the development of instance-reduction techniques aimed at retaining essential instances and discarding redundant ones. While such trimming optimizes computational demands, it might adversely affect classification accuracy. This study introduces the novel Selective Learning Vector Quantization (SLVQ) algorithm, specifically designed to enhance the performance of datasets reduced through such techniques. Unlike traditional LVQ algorithms that employ random vector weights (codebook vectors), SLVQ utilizes instances selected by the reduction algorithm as the initial weight vectors. Importantly, as these instances often contain nominal values, SLVQ modifies the distances between these nominal values, rather than modifying the values themselves, aiming to improve their representation of the training set. This approach is crucial because nominal attributes are common in real-world datasets and require effective distance measures, such as the Value Difference Measure (VDM), to handle them properly. Therefore, SLVQ adjusts the VDM distances between nominal values, instead of altering the attribute values of the codebook vectors. Hence, the innovation of the SLVQ approach lies in its integration of instance reduction techniques for selecting initial codebook vectors and its effective handling of nominal attributes. Our experiments, conducted on 17 text classification datasets with four different instance reduction algorithms, confirm SLVQ’s effectiveness. It significantly enhances the kNN’s classification accuracy of reduced datasets. In our empirical study, the SLVQ method improved the performance of these datasets, achieving average classification accuracies of 82.55%, 84.07%, 78.54%, and 83.18%, compared to the average accuracies of 76.25%, 79.62%, 66.54%, and 78.19% achieved by non-fine-tuned datasets, respectively.

Ensemble LVQ Model for Photovoltaic Line-to-Line Fault Diagnosis Using K-Means Clustering and AdaGrad

Line-to-line (LL) faults are one of the most frequent short-circuit conditions in photovoltaic (PV) arrays which are conventionally detected and cleared by overcurrent protection devices (OCPDs). However, OCPDs are shown to face challenges when detecting LL faults under critical detection conditions, i.e., low mismatch levels and/or high fault impedance values. This occurs due to insufficient fault current, thus leaving the LL faults undetected and leading to power losses and even catastrophic fire hazards. To compensate for OCPD deficiencies, recent studies have proposed modern artificial intelligence (AI)-based methods. However, various limitations can still be witnessed even in AI-based methods, such as (i) most of the models requiring a massive training dataset, (ii) critical fault detection conditions not being taken into consideration, (iii) models not being accurate enough when dealing with critical conditions, etc. To this end, the present paper proposes a learning vector quantization (LVQ)-based ensemble learning model in which three LVQs are individually trained to detect and classify LL faults in PV arrays. The initial LVQ vectors are determined using the k-means clustering method, and the learning rate is optimized by the adaptive gradient (AdaGrad) optimizer. The training and testing datasets are collected according to the PV array’s current–voltage (I–V) characteristic curve, and several features are extracted based on the Canberra and chi-squared distance techniques. The model utilizes a small training dataset, considers various critical detection conditions for LL faults—such as different mismatch levels and fault impedance values—and the final experimental results show that the model achieves an impressive average accuracy of 99.26%.

SVQ-MAE: an efficient speech pre-training framework with constrained computational resources

Self-supervised learning for speech pre-training models has achieved remarkable success in acquiring superior speech contextual representations by learning from unlabeled audio, excelling in numerous downstream speech tasks. However, the pre-training of these models necessitates significant computational resources and training duration, presenting a high barrier to entry into the realm of pre-training learning. In our efforts, by amalgamating the resource-efficient benefits of the generative learning model, Masked Auto Encoder, with the efficacy of the vector quantization method in discriminative learning, we introduce a novel pre-training framework: Speech Vector Quantization Masked Auto Encoder (SVQ-MAE). Distinct from the majority of SSL frameworks, which require simultaneous construction of speech contextual representations and mask reconstruction within an encoder-only module, we have exclusively designed a decoupled decoder for pre-training SVQ-MAE. This allows the additional decoupled decoder to undertake the mask reconstruction task solely, reducing the learning complexity of pretext tasks and enhancing the encoder’s efficiency in extracting speech contextual representations. Owing to this innovation, by using only 4 GPUs, SVQ-NAE can achieve high performance comparable to wav2vec 2.0, which requires 64 GPUs for training. In the Speech Processing Universal Performance Benchmark, SVQ-MAE surpasses wav2vec 2.0 in both keyword spotting and emotion recognition tasks. Furthermore, in cross-lingual ASR for Mandarin, upon fine-tuning on AISHELL-1, SVQ-MAE achieves a Character Error Rate of 4.09%, outperforming all supervised ASR models.

Machine learning-based identification and validation of immune-related biomarkers for early diagnosis and targeted therapy in diabetic retinopathy

The early diagnosis of diabetic retinopathy (DR) is challenging, highlighting the urgent need to identify new biomarkers. Immune responses play a crucial role in DR, yet there are currently no reports of machine learning (ML) algorithms being utilized for the development of immune-related molecular markers in DR. Based on the datasets GSE102485 and GSE160306, differentially expressed genes (DEGs) were screened using Weighted Gene Co-expression Network Analysis (WGCNA). Five ML algorithms including Bayesian, Learning Vector Quantization (LVQ), Wrapper (Boruta), Random Forest (RF), and Logistic Regression were employed to select immune-related genes associated with DR (DR.Sig). Seven ML algorithms including Naive Bayes (NB), RF, Support Vector Machine (SVM), AdaBoost Classification Trees (AdaBoost), Boosted Logistic Regressions (LogitBoost), K-Nearest Neighbors (KNN), and Cancerclass were utilized to construct a predictive model for DR. The relationship between DR.Sig genes and immune cells was analyzed using single-sample Gene Set Enrichment Analysis (ssGSEA). Additionally, drug sensitivity prediction of DR.Sig genes and molecular docking were performed. Through the utilization of 5 ML algorithms, 6 immune-related biomarkers closely related to the occurrence of DR were identified, including FCGR2B, CSRP1, EDNRA, SDC2, TEK, and CIITA. The DR predictive model constructed based on these 6 DR.Sig genes using the Cancerclass algorithm demonstrated superior predictive performance compared to 4 previously published DR-related biomarkers. In vivo and in vitro experiments also provided strong validation of the expression of the 6 genes in DR. Positive correlations were observed between these genes and 22 types of immune cells. Molecular docking results revealed that CSRP1, EDNRA, and TEK exhibited the highest affinities with the small molecule compounds etoposide, FR-139317, and camptothecin, respectively. The models constructed based on various ML algorithms can effectively predict the occurrence of DR events and hold potential for targeted drug therapies, providing a basis for the early diagnosis and targeted treatment of DR.

An efficient infinity norm minimization algorithm for under-determined inverse problems

The problem of solving under-determined systems of linear equations with minimum peak magnitude (ℓ∞ norm) has numerous applications in signal processing. These include Peak-to-Average Power Ratio (PAPR) reduction in MIMO-OFDM systems, vector quantization, approximate nearest neighbor search, optimal control in robotics, and power grid optimization. Several methods have been proposed to address this problem, but they often face limitations in computational speed or representation accuracy. Some methods also impose constraints on the frame matrix, such as restrictions on the type of its entries or its aspect ratio. In this paper, we present the Fast Iterative Peak Shrinkage Algorithm (FIPSA), which iterates over feasible solutions to consistently reduce peak magnitude and provably converge to near-optimal solutions. Our experimental results, conducted across various frame matrix types and aspect ratios, demonstrate that FIPSA consistently achieves near-minimal ℓ∞ norm values. In addition, it operates 1.3 to 7.3 times faster than previous methods, while maintaining an average representation error of 10−15. Notably, these advancements are achieved without imposing any constraints on the frame matrix.

Quantization‐Based Latin Hypercube Sampling for Dependent Inputs With an Application to Sensitivity Analysis of Environmental Models

ABSTRACTNumerical models are essential for comprehending intricate physical phenomena in different domains. To handle their complexity, sensitivity analysis, particularly screening is crucial for identifying influential input parameters. Kernel‐based methods, such as the Hilbert‐Schmidt Independence Criterion (HSIC), are valuable for analyzing dependencies between inputs and outputs. Implementing HSIC requires data from the original model, which leads to the need of efficient sampling strategies to limit the number of costly numerical simulations. While, for independent input variables, existing sampling methods like Latin Hypercube Sampling (LHS) are effective in estimating HSIC with reduced variance, incorporating dependence is challenging. This article introduces a novel LHS variant, quantization‐based LHS (QLHS), which leverages Voronoi vector quantization to address dependent inputs. The method provides good coverage of the range of variations in the input variables. The article outlines expectation estimators based on QLHS in various dependency settings, demonstrating their unbiasedness. The method is applied to several models of growing complexities, first on simple examples to illustrate the theory, then on more complex environmental hydrological models, when the dependence is known or not, and with more and more interactive processes and factors. The last application is on the digital twin of a French vineyard catchment (Beaujolais region) to design a vegetative filter strip and reduce water, sediment, and pesticide transfers from the fields to the river. QLHS is used to compute HSIC measures and independence tests, demonstrating its usefulness, especially in the context of complex models.

Vector quantization-driven image compression through multi-objective evolutionary algorithms

In recent years, information consumption among the population has increased, especially through the widespread use of digital images and videos. Consequently, the demand for efficient compression methods has escalated, aiming to facilitate both data storage and transmission while adapting to the unique requirements of diverse problems. The image compression problem has been explored in several works through the lens of vector quantization (VQ) by adopting a single-objective approach to achieve optimal quality results but with a fixed compression level. In contrast, this investigation introduces a groundbreaking multi-objective compression approach that simultaneously optimizes image quality and compression level. This is achieved by selecting the most representative information to construct an adequate codebook while minimizing its size. The proposed method was evaluated on a curated selection of well-known public domain images, using different types of multi-objective evolutionary algorithms, including Pareto-, indicator- and decomposition-based approaches. The compressed images produced were assessed through the application of established quality indicators drawn from the expansive realm of image processing literature, ensuring the reliability and validity of our findings. The results demonstrate the proposed method’s ability to adapt to the intrinsic characteristics of each image, removing the maximum amount of redundant information presented. Furthermore, the multi-objective nature of the proposed approach allows us to obtain solutions with different levels of trade-off between quality and compression, opening up new possibilities in the field and demonstrating the potential to address the evolving challenges posed by the escalating volume of information in the digital landscape.

Event-triggered and quantised feedback stabilisation of switched networked control systems

In order to conserve network resources, this paper investigates the event-triggered control and quantised feedback stabilisation of switched networked control systems by introducing the vector quantizer technique. The state information undergoes quantisation and is transmitted to the controller after being sampled by the event-triggering technology. The designed event-triggered mechanism relies on quantised state values, and the generation of each quantised value also depends on the event-triggered condition. Addressing the asynchronous issue between the controller and the subsystem, we provide a sufficient condition for ensuring exponential stability of the closed-loop system and outline the controller gain design method by employing the multiple Lyapunov function and average dwell time methods. Besides, we demonstrate the explicit exclusion of the Zeno phenomenon in the event-triggered mechanism. Finally, the viability of our approach is demonstrated through simulation examples.

Quantized kernel recursive q-Rényi-like algorithm

This paper introduces the kernel recursive q-Rényi-like (KRqRL) algorithm, based on the q-Rényi kernel function and the kernel recursive least squares (KRLS) algorithm. To reduce the computational complexity and memory requirements of the KRqRL algorithm, an online vector quantization (VQ) method is employed to quantize the network size to a codebook size, resulting in the quantized KRqRL (QKRqRL) algorithm. This paper provides a detailed analysis of the convergence and computational complexity of the QKRqRL algorithm. In the simulation experiments, the network size of each algorithm is reduced to 25% of its original size. The performance of the QKRqRL algorithm is evaluated in terms of convergence speed, prediction error, and computation time under non-Gaussian noise conditions. Finally, the QKRqRL algorithm is further validated using sunspot data, demonstrating its superior stability and online prediction performance.

Prediction of surface roughness at the machining of densified wood surfaces

ABSTRACT In this research, modeling and interpretation of the effects of the machining parameters on surface roughness (Rz) using artificial neural networks (ANN) and analysis methods (MLP, LVQ and Taguchi Design analysis) werw studied. Black poplar (Populus nigra) was selected as the experimental material. Specimens, which were transversal densified by a thermo-mechanical (TM) process at 0%, 20% and 40% compression ratios. The densified wood was processed at 1000, 1500 and 2000 mm/min feed speeds and in 12,000, 15,000, and 18,000 rpm rotation speeds in a CNC machine by using two different cutters. Rz surface-roughness values were measured. The modeling of Rz with artificial neural networks (ANN) is discussed. While 97% success was achieved in the classification made with multilayer perceptron (MLP) in estimating Rz, this success rate is 100% in the classification made with learning vector quantization (LVQ). MLP and LVQ are operations in the MATLAB R2019b software. Taguchi values were compared with the experimental data results, it was seen that the MSE value was 4.6%, the MAE value was 1.8%, and the RMSE value was 2.1%. These results showed that the presented models can accurately determine the appropriate parameters for the desired Rz. The prediction results and presented models can be applied for targeted industrial applications.

Adaptive Neighbors Graph Learning for Large-Scale Data Clustering using Vector Quantization and Self-Regularization

In traditional adaptive neighbors graph learning (ANGL)-based clustering, the time complexity is more than O(n2), where n is the number of data points, which is not scalable for large-scale data problems in real applications. Subsequently, ANGL adds a balance regularization to its objective function to avoid the sparse over-fitting problem in the learned similarity graph matrix. Still, the regularization may leads to many weak connections between data points in different clusters. To address these problems, we propose a new fast clustering method, namely, Adaptive Neighbors Graph Learning for Large-Scale Data Clustering using Vector Quantization and Self-Regularization (ANGL-LDC), to perform vector quantization (VQ) on original data and feed the obtained VQ data as the input in the n×n similarity graph matrix learning. Hence, the n×n similarity graph matrix learning problem is simplified to weighted m×m(m≪n) graph learning problem, where m is the number of distinct points and weight is the duplicate times of distinct points in VQ data. Consequently, the time complexity of ANGL-LDC is much lower than that of ANGL. At the same time, we propose a new ANGL objective function with a graph connection self-regularization mechanism, where the ANGL-LDC objective function will get an infinity value if the value of one graph connection is equal to 1. Therefore, ANGL-LDC naturally avoids obtaining the sparse over-fitting problem since we need to minimize the value of ANGL-LDC’s objective function. Experimental results on synthetic and real-world datasets demonstrate the scalability and effectiveness of ANGL-LDC.

Breast Tumor Diagnosis Based on Molecular Learning Vector Quantization Neural Networks.

DNA nanotechnology plays a crucial role in precise cancer medicine. Currently, molecular logic circuits are applied to detect tumor-specific biomarkers and control the release of therapeutic drugs. However, these systems lack self-learning capabilities for intelligent diagnostics in biological samples, and their data processing capabilities are limited. Here, a molecular learning vector quantization neural network (LVQNN) model based on DNA strand displacement (DSD) technology for breast tumor diagnosis is developed. Compared to previous work, the molecular LVQNN boasts powerful computing abilities, handling high-dimensional data for intelligent cancer diagnosis. To verify the feasibility and versatility of the network, two distinct typical datasets are selected: one from a single source with cell morphology data from 569 cases, and a more extensive one spanning different populations and ages, with miRNA gene expression data from 1881 cases. By using the molecular LVQNN, diagnostic experiments are conducted on 50 and 120 public individuals from these two datasets, respectively, achieving accuracy rates of 94% and 97.5%. This study demonstrates that the LVQNN model exhibits significant advantages in breast cancer diagnosis and enhances diagnostic accuracy while introducing new approaches for intelligent cancer diagnosis, anticipated to bring significant breakthroughs and application prospects to precise cancermedicine.

Systematization of sunflower genotypes based on seed phenotypic characteristics using neural networks

In recent years, various datasets related to the phenotyping of sunflower genotypes have become increasingly accessible. However, one of the key challenges remains the efficient and accurate prediction of phenotypes based on genotypes in the context of climate change. Analyzing phenotypes at different levels of organization and detecting connections between phenotypes and genotypes require the integration and processing of large, diverse, and often noisy datasets. Machine learning offers a broad arsenal of methods and approaches for identifying predictive patterns in such data. Therefore, the research aimed to develop a methodology for the systematization of sunflower genotypes based on seed phenotypic characteristics using the data vector quantization method and neural networks. The study revealed the phenotypic characteristics of sunflower seeds from various genotypes selected by the Institute of Oilseed Crops of NAAS, grown in the southern Steppe of Ukraine, including seed length, width, thickness, seed mass, kernel mass, and seed coat cracking force. For this purpose, appropriate laboratory equipment was developed, including two modules for determining the morphological and rheological properties of seeds. The developed methodology for the systematization of sunflower genotypes based on seed phenotypic characteristics includes the following steps: measuring the characteristics of sunflower seeds from various samples (parental components); studying the mutual correlation of characteristics; conducting hierarchical cluster analysis of the data using the Ward's method; determining the optimal number of groups; performing k-means clustering using the vector quantization method; determining the correspondence of ranges of characteristics to the group; training a neural network to group the data by samples and created groups; verifying the adequacy of the neural network on test data. The developed methodology was tested, and the MLP 30-15-3 neural network for grouping data by samples and created groups of sunflower seeds was developed in the Statistica software package. The network's training efficiency was 99.4%, and such of testing and validation was 95.6% and 96.7%, respectively.

One-index vector quantization based adversarial attack on image classification

To improve storage and transmission, images are generally compressed. Vector quantization (VQ) is a popular compression method as it has a high compression ratio that suppresses other compression techniques. Despite this, existing adversarial attack methods on image classification are mostly performed in the pixel domain with few exceptions in the compressed domain, making them less applicable in real-world scenarios. In this paper, we propose a novel one-index attack method in the VQ domain to generate adversarial images by a differential evolution algorithm, successfully resulting in image misclassification in victim models. The one-index attack method modifies a single index in the compressed data stream so that the decompressed image is misclassified. It only needs to modify a single VQ index to realize an attack, which limits the number of perturbed indexes. The proposed method belongs to a semi-black-box attack, which is more in line with the actual attack scenario. We apply our method to attack three popular image classification models, i.e., Resnet, NIN, and VGG16. On average, 55.9 % and 77.4 % of the images in CIFAR-10 and Fashion MNIST, respectively, are successfully attacked, with a high level of misclassification confidence and a low level of image perturbation.

VQ-NeRF: Neural Reflectance Decomposition and Editing With Vector Quantization.

We propose VQ-NeRF, a two-branch neural network model that incorporates Vector Quantization (VQ) to decompose and edit reflectance fields in 3D scenes. Conventional neural reflectance fields use only continuous representations to model 3D scenes, despite the fact that objects are typically composed of discrete materials in reality. This lack of discretization can result in noisy material decomposition and complicated material editing. To address these limitations, our model consists of a continuous branch and a discrete branch. The continuous branch follows the conventional pipeline to predict decomposed materials, while the discrete branch uses the VQ mechanism to quantize continuous materials into individual ones. By discretizing the materials, our model can reduce noise in the decomposition process and generate a segmentation map of discrete materials. Specific materials can be easily selected for further editing by clicking on the corresponding area of the segmentation outcomes. Additionally, we propose a dropout-based VQ codeword ranking strategy to predict the number of materials in a scene, which reduces redundancy in the material segmentation process. To improve usability, we also develop an interactive interface to further assist material editing. We evaluate our model on both computer-generated and real-world scenes, demonstrating its superior performance. To the best of our knowledge, our model is the first to enable discrete material editing in 3D scenes.

A maritime image dehazing algorithm based on improved VQGAN

ABSTRACT Maritime image dehazing faces challenges as existing methods struggle to handle different concentrations of sea fog, and the treatment of details in dehazed results is often suboptimal. This paper proposes a maritime image dehazing algorithm based on the VQGAN mode. Specifically, since VQGAN is capable of generating high-quality discrete codebooks, we input the haze-free maritime dataset into VQGAN for pre-training to obtain a high-quality codebook prior. During the vector quantisation stage,high-quality vectors from the discrete codebook undergo distance calculations with the upsampled vectors containing haze.This process replaces the vectors containing haze with the high-quality vectors. To address the issue of residual fog, the authors introducing the FAM-UNet encoder.To tackle distortion issues, the authors introduced the Pyramid, Cascading, and Deformable Convolution Feature Alignment Module (PCD) into the original model. The experimental results indicate that our model can effectively restore clear images, providing assistance for subsequent advanced computer vision tasks..