Evaluating shape and appearance descriptors for 3D human pose estimation

3D human pose estimation aims to reconstruct the human skeleton of all the individuals in a scene by detecting several body joints. The creation of accurate and efficient methods is required for several real-world applications including animation, human–robot interaction, surveillance systems or sports, among many others. However, several obstacles such as occlusions, random camera perspectives, or the scarcity of 3D labelled data, have been hampering the models’ performance and limiting their deployment in real-world scenarios. The higher availability of cameras has led researchers to explore multi-view solutions due to the advantage of being able to exploit different perspectives to reconstruct the pose. Most existing reviews focus mainly on monocular 3D human pose estimation and a comprehensive survey only on multi-view approaches to determine the 3D pose has been missing since 2012. Thus, the goal of this survey is to fill that gap and present an overview of the methodologies related to 3D pose estimation in multi-view settings, understand what were the strategies found to address the various challenges and also, identify their limitations. According to the reviewed articles, it was possible to find that most methods are fully-supervised approaches based on geometric constraints. Nonetheless, most of the methods suffer from 2D pose mismatches, to which the incorporation of temporal consistency and depth information have been suggested to reduce the impact of this limitation, besides working directly with 3D features can completely surpass this problem but at the expense of higher computational complexity. Models with lower supervision levels were identified to overcome some of the issues related to 3D pose, particularly the scarcity of labelled datasets. Therefore, no method is yet capable of solving all the challenges associated with the reconstruction of the 3D pose. Due to the existing trade-off between complexity and performance, the best method depends on the application scenario. Therefore, further research is still required to develop an approach capable of quickly inferring a highly accurate 3D pose with bearable computation cost. To this goal, techniques such as active learning, methods that learn with a low level of supervision, the incorporation of temporal consistency, view selection, estimation of depth information and multi-modal approaches might be interesting strategies to keep in mind when developing a new methodology to solve this task. • First review only on multi-view, multi-modal methods to estimate 3D pose since 2012. • Multi-view allows capturing the full body geometry, making 3D pose estimation easier. • Real-world applications include sports, broadcasting, rehabilitation or animation. • Finding a fast, accurate method with low computational cost remains a challenge. • Multi-modal methods or view selection can lead to an efficient and effective model.

Although Relevance Vector Machine (RVM) is the most popular algorithms in machine learning and computer vision, outliers in the training data make the estimation unreliable. In the paper, a robust RVM model under non-parametric Bayesian framework is proposed. We decompose the noise term in the RVM model into two components, a Gaussian noise term and a spiky noise term. Therefore the observed data is assumed represented as: y= Dw+s+e where Dw is the relevance vector component, of which D is the kernel function matrix and w is the weight matrix, s is the spiky term and e is the Gaussian noise term. A spike-slab sparse prior is imposed on the weight vector, w which gives a more intuitive constraint on the sparsity than the Student’s t-distribution described in the traditional RVM. For the spiky component, s a spike-slab sparse prior is also introduced to recognize outliers in the training data effectively. Several experiments demonstrate the better performance over the RVM regression.

Existed dengue prediction model associated with temperature data are always based on Poisson regression methods or linear models. However, these models are difficult to be applied to non-stationary climate data, such as rainfall or precipitation. A novel k-nearest neighbor (KNN) regression method was proposed to improve the prediction accuracy of dengue fever regression model in this paper. The dengue cases and the climatic factors (average minimum temperature, average maximum temperature, average temperature, average dew point temperature, temperature difference, relative humidity, absolute humidity, Precipitation) in San Juan, Puerto Rico during the period 1990–2013 were regressed by the KNN algorithm. The performances of KNN regression were studied by compared with correlation analysis and Poisson regression method. Results showed that the KNN model fitted real dengue outbreak better than Poisson regression method while the root mean square error was 6.88.

Weakly-supervised pre-training for 3D human pose estimation via perspective knowledge

Application of relevance vector machine and logistic regression for machine degradation assessment

Machines learning models have recently been proposed for predicting rivers water temperature (Tw) using only air temperature (Ta). The proposed models relied on a nonlinear relationship between the Tw and Ta and they have proven to be robust modelling tools. The main motivation for this study was to evaluate how the variational mode decomposition (VMD) contributed to the improvement of machines learning performances for river Tw modelling. Measured data collected at five stations located in Poland from 1987 to 2014 were acquired and used for the analysis. Six machines learning models were used and compared namely, K-nearest neighbor's regression (KNNR), least square support vector machine (LSSVM), generalized regression neural network (GRNN), cascade correlation artificial neural networks (CCNN), relevance vector machine (RVM), and locally weighted polynomials regression (LWPR). The six models were developed according to three scenarios. First, the models were calibrated using only the Ta as input and obtained results show that the models were able to predict consistently water temperature, showing a high determination coefficient (R2) and Nash-Sutcliffe efficiency (NSE) with values near or above 0.910 and 0.915, respectively, and in overall the six models worked equally without clear superiority of one above another. Second, the air temperature was combined with the periodicity (i.e., day, month and year number) as input variable and a significant improvement was achieved. Both models show their ability to accurately predict river Tw with an overall accuracy of 0.956 for R2 and 0.955 for NSE values, but the LSSVM2 have some advantages such as a small errors metrics, and high fitting capabilities and it slightly surpasses the others models. Thirdly, air temperature was decomposed into several intrinsic mode functions (IMF) using the VMD method and the performances of the models were evaluated. The VMD parameters appeared to cause much influence on the prediction accuracy, exhibiting an improvement of about 40.50% and 39.12% in terms of RMSE and MAE between the first and the third scenarios, however, some models, i.e., GRNN and KNNR have not benefited from the VMD. This research has demonstrated the high capability of the VMD algorithm as a preprocessing approach in improving the accuracies of the machine learning models for river water temperature prediction.

Laser-induced Breakdown spectroscopy quantitative analysis method via adaptive analytical line selection and relevance vector machine regression model

Due to the influence of many factors, there is no reasonable evaluation method for the heat loss evaluation of the envelope, which leads to the deviation of the evaluation results of building energy consumption. By comparing different regression analysis models and considering the heat transfer characteristics of the envelope and heat loss evaluation needs, the dynamic reference value of heat loss evaluation “temperature baseline” which meets different energy saving standards is established. The evaluation method of instantaneous heat loss of the envelope is proposed and the heat loss is studied. The results showed that the performance indexes of relevance vector machine (RVM) regression model was superior to the other two regression models(response surface methodology (RSM), support vector machine (SVM). Based on the RVM regression model, the dynamic prediction model of “temperature baseline” was established by combining numerical simulation with regression analysis. Taking a residential building as an example, the envelope temperature baselines meeting the energy saving standard of 75% and 50% were obtained by using the prediction model, the temperature baselines of walls and windows that save 75% energy were −0.6°C and 1.9 °C respectively, and those that save 50% energy were 1.2°C and 4.5°C. The infrared image and heat loss evaluation method found that 87% of the points exceeded the temperature baseline by more than 2°C, and the total heat loss of the wall with an area of 13m2 was 471W. The heat loss evaluation method based on RVM proposed in this paper can objectively evaluate the heat loss of building.

Deep 3D human pose estimation: A review

Generalization is very important for pose estimation, especially for 3D pose estimation where small changes in the 2D images could trigger structural changes in the 3D space. To achieve generalization, the system needs to have the capability of detecting estimation errors by double-checking the projection coherence between the 3D and 2D spaces and adapting its network inference process based on this feedback. Current pose estimation is one-time feed-forward and lacks the capability to gather feedback and adapt the inference outcome. To address this problem, we propose to explore the concept of progressive inference where the network learns an observer to continuously detect the prediction error based on constraints matching, as well as an adjuster to refine its inference outcome based on these constraints errors. Within the context of 3D hand pose estimation, we find that this observer-adjuster design is relatively unstable since the observer is operating in the 2D image domain while the adjuster is operating in the 3D domain. To address this issue, we propose to construct two sets of observers-adjusters with complementary constraints from different perspectives. They operate in a dynamic sequential manner controlled by a decision network to progressively improve the 3D pose estimation. We refer to this method as Cross-Constrained Progressive Inference (CCPI). Our extensive experimental results on FreiHAND and HO-3D benchmark datasets demonstrate that the proposed CCPI method is able to significantly improve the generalization capability and performance of 3D hand pose estimation.

In petroleum geophysics, well logs are fundamental for subsurface characterization; however, missing logs frequently occur due to tool failure, legacy data gaps, or economic constraints, limiting reliable reservoir evaluation. The primary aim of this study is to develop and evaluate a simple, nonparametric machine learning framework for predicting missing geophysical well logs using K-Nearest Neighbors (KNN) regression. A secondary aim is to assess the effectiveness of correlation-guided feature selection and distance-based learning for estimating continuous log responses in data-limited scenarios. The proposed workflow is applied to an open-source dataset from the University of Kansas and focuses on predicting Sonic (Delta T, DT) and Gamma Ray (GR) logs across five wells. Pearson correlation analysis is used to identify the most relevant input features, followed by min–max normalization to ensure distance metric consistency. The dataset is split into training and testing subsets using a 70/30 ratio, and the optimal number of neighbors (k) is determined by minimizing the root mean square error (RMSE). The KNN models achieved high predictive performance, with test R 2 values ranging from 0.942 to 0.963 for DT logs and from 0.927 to 0.930 for GR logs. Although minor overfitting was observed, the difference between training and testing performance remained limited, indicating satisfactory generalization. The results demonstrate that KNN regression provides a robust and computationally efficient solution for missing well-log prediction in geophysical applications, offering a practical alternative for enhancing reservoir characterization when log data are incomplete.

Recently, the emergence of Kinect systems has demonstrated the benefits of predicting an intermediate body part labeling for 3D human pose estimation, in conjunction with RGB-D imagery. The availability of depth information plays a critical role, so an important question is whether a similar representation can be developed with sufficient robustness in order to estimate 3D pose from RGB images. This paper provides evidence for a positive answer, by leveraging (a) 2D human body part labeling in images, (b) second-order label-sensitive pooling over dynamically computed regions resulting from a hierarchical decomposition of the body, and (c) iterative structured-output modeling to contextualize the process based on 3D pose estimates. For robustness and generalization, we take advantage of a recent large-scale 3D human motion capture dataset, Human3.6M[18] that also has human body part labeling annotations available with images. We provide extensive experimental studies where alternative intermediate representations are compared and report a substantial 33% error reduction over competitive discriminative baselines that regress 3D human pose against global HOG features.

Plants play a crucial role in foodstuff, medicine, industry, and environmental protection. The skill of recognising plants is very important in some applications, including conservation of endangered species and rehabilitation of lands after mining activities. However, it is a difficult task to identify plant species because it requires specialized knowledge. Developing an automated classification system for plant species is necessary and valuable since it can help specialists as well as the public in identifying plant species easily. Shape descriptors were applied on the myDAUN dataset that contains 45 tropical shrub species collected from the University of Malaya (UM), Malaysia. Based on literature review, this is the first study in the development of tropical shrub species image dataset and classification using a hybrid of leaf shape and machine learning approach. Four types of shape descriptors were used in this study namely morphological shape descriptors (MSD), Histogram of Oriented Gradients (HOG), Hu invariant moments (Hu) and Zernike moments (ZM). Single descriptor, as well as the combination of hybrid descriptors were tested and compared. The tropical shrub species are classified using six different classifiers, which are artificial neural network (ANN), random forest (RF), support vector machine (SVM), k-nearest neighbour (k-NN), linear discriminant analysis (LDA) and directed acyclic graph multiclass least squares twin support vector machine (DAG MLSTSVM). In addition, three types of feature selection methods were tested in the myDAUN dataset, Relief, Correlation-based feature selection (CFS) and Pearson’s coefficient correlation (PCC). The well-known Flavia dataset and Swedish Leaf dataset were used as the validation dataset on the proposed methods. The results showed that the hybrid of all descriptors of ANN outperformed the other classifiers with an average classification accuracy of 98.23% for the myDAUN dataset, 95.25% for the Flavia dataset and 99.89% for the Swedish Leaf dataset. In addition, the Relief feature selection method achieved the highest classification accuracy of 98.13% after 80 (or 60%) of the original features were reduced, from 133 to 53 descriptors in the myDAUN dataset with the reduction in computational time. Subsequently, the hybridisation of four descriptors gave the best results compared to others. It is proven that the combination MSD and HOG were good enough for tropical shrubs species classification. Hu and ZM descriptors also improved the accuracy in tropical shrubs species classification in terms of invariant to translation, rotation and scale. ANN outperformed the others for tropical shrub species classification in this study. Feature selection methods can be used in the classification of tropical shrub species, as the comparable results could be obtained with the reduced descriptors and reduced in computational time and cost.

In the field of 3D Human Pose Estimation and Reconstruction based on body joints extracted from a 2D image; Exists challenges like self-occlusion and depth perception. These problems hinder approximate estimations. This article proposes a hybrid method that consists of semantic segmentation, sparse representation and 2D pose estimation for 3D estimation. Specifically, we train two fully Convolution neural networks (FCNs) to estimate 2D pose and semantic segmentation. Then, we narrow the estimation down using basic human body structure and results from the FCNs. Subsequently, utilizing the estimated 2D pose of the earlier step and a sparse representation model the 3D pose will estimate. Using a Convolution Neural Networks learning method for 2D human pose estimation and then by sparse representation will estimate. Results of the proposed method, show improvement relative to the previous works on 3D Human pose estimation. 3D Human Pose Estimation by this method demonstrates a significant mean error reduction compared to earlier studies.

It is widely shown that Convolutional Neural Networks (CNNs) are vulnerable to adversarial examples on most recognition tasks, such as image classification and segmentation. However, few work studies the more complicated task – 3D human pose estimation. This task often requires large-scale datasets, specialized network architectures, and it can be solved either from single-view RGB images or from multi-view RGB images. In this paper, we make the first attempt to investigate the robustness of current state-of-the-art 3D human pose estimation methods. To this end, we build four representative baseline models, where most of the current methods can be generally classified as one of them. Furthermore, we design targeted adversarial attacks to detect whether 3D pose estimators are robust to different camera parameters. For different types of methods, we present a comprehensive study of their robustness on the large-scale Human3.6M benchmark. Our work shows that different methods vary significantly in their resistance to adversarial attacks. Through extensive experiments, we show that multi-view 3D pose estimators can be more vulnerable to adversarial examples. We believe that our efforts can shed light on future works to design more robust 3D human pose estimators.