Part Of Representation Research Articles

Unsupervised Part Discovery via Dual Representation Alignment.

Object parts serve as crucial intermediate representations in various downstream tasks, but part-level representation learning still has not received as much attention as other vision tasks. Previous research has established that Vision Transformer can learn instance-level attention without labels, extracting high-quality instance-level representations for boosting downstream tasks. In this paper, we achieve unsupervised part-specific attention learning using a novel paradigm and further employ the part representations to improve part discovery performance. Specifically, paired images are generated from the same image with different geometric transformations, and multiple part representations are extracted from these paired images using a novel module, named PartFormer. These part representations from the paired images are then exchanged to improve geometric transformation invariance. Subsequently, the part representations are aligned with the feature map extracted by a feature map encoder, achieving high similarity with the pixel representations of the corresponding part regions and low similarity in irrelevant regions. Finally, the geometric and semantic constraints are applied to the part representations through the intermediate results in alignment for part-specific attention learning, encouraging the PartFormer to focus locally and the part representations to explicitly include the information of the corresponding parts. Moreover, the aligned part representations can further serve as a series of reliable detectors in the testing phase, predicting pixel masks for part discovery. Extensive experiments are carried out on four widely used datasets, and our results demonstrate that the proposed method achieves competitive performance and robustness due to its part-specific attention.

Exploring Smokers Perspectives on Health Status and Health Seeking Behavior: A Qualitative Exploration

Objective: Smokers might tend to ignore when communicating about smoking health risks. We used Bourdieu’s conceptual tool to guide and explore their perception of being healthy despite their smoking behavior and their health seeking practice. Material and Methods: A qualitative method, using a semi-structured interview as the main method of collecting the data, was used. This method was implemented because of its flexibility in the exploration, based on participant responses. Results: The findings are reported as per the thematic analysis of 15 purposively selected participants. The ages of the 15 participants varied from 20 to 55 years old, and they were selected from the quantitative study to participate voluntarily in this qualitative exploration. Among them, 14 were males and one was female. The ethnic representation in this qualitative part were Malays (n=14) and Chinese (n=1). The duration of smoking varied in the sample, with the majority smoking for more than five years. Conclusion: This study better understands smokers’ health perception and their health-seeking behavior. Firstly, the perception of the less harmful effects of smoking on the health of smokers was understood within the views of Bourdieu’s Theory of Cultural Health Capital, which, in turn poses challenges to smoking cessation programs and campaigns. This study reported that smokers perceived themselves as healthy, and thus continued their smoking habit despite knowing its side effects.

ANISE: Assembly-Based Neural Implicit Surface Reconstruction.

We present ANISE, a method that reconstructs a 3D shape from partial observations (images or sparse point clouds) using a part-aware neural implicit shape representation. The shape is formulated as an assembly of neural implicit functions, each representing a different part instance. In contrast to previous approaches, the prediction of this representation proceeds in a coarse-to-fine manner. Our model first reconstructs a structural arrangement of the shape in the form of geometric transformations of its part instances. Conditioned on them, the model predicts part latent codes encoding their surface geometry. Reconstructions can be obtained in two ways: (i) by directly decoding the part latent codes to part implicit functions, then combining them into the final shape; or (ii) by using part latents to retrieve similar part instances in a part database and assembling them in a single shape. We demonstrate that, when performing reconstruction by decoding part representations into implicit functions, our method achieves state-of-the-art part-aware reconstruction results from both images and sparse point clouds. When reconstructing shapes by assembling parts retrieved from a dataset, our approach significantly outperforms traditional shape retrieval methods even when significantly restricting the database size. We present our results in well-known sparse point cloud reconstruction and single-view reconstruction benchmarks.

Scalable and efficient digital twins for model-based design of cyber-physical systems

ABSTRACT The optimised design, operation and management of complex, large-size Cyber Physical Systems (CPSs) – like modern manufacturing and logistic assets – calls for Digital Twins (DTs) in which dynamic modelling and simulation play a relevant role. In such simulation-based DTs computational efficiency is crucial. According to the present state of the art, the said efficiency is hindered by the way the Cyber part of a CPS is represented, which in the manufacturing case practically corresponds to representing digital controls. This paper proposes a modelling framework for an efficient and scalable representation of the Cyber part in a DT of a CPS. The presented solution is based on Object-Oriented Modelling (OOM) and IEC industrial standards. Motivating and explanatory examples are provided, and a proof-of-concept case study is discussed to support the framework potential and industrial viability.

Quantum simulation of dissipation for Maxwell equations in dispersive media

The dissipative character of an electromagnetic medium breaks the unitary evolution structure that is present in lossless, dispersive optical media. In dispersive media, dissipation appears in the Schrödinger representation of classical Maxwell equations as a sparse diagonal operator occupying an r-dimensional subspace. A first order Suzuki-Trotter approximation for the evolution operator enables us to isolate the non-unitary operators (associated with dissipation) from the unitary operators (associated with lossless media). The unitary operators can be implemented through qubit lattice algorithm (QLA) on n qubits, based on the discretization and the dimensionality of the pertinent fields. However, the non-unitary-dissipative part poses a challenge both physically and computationally on how it should be implemented on a quantum computer. In this paper, two probabilistic dilation algorithms are considered for handling the dissipative operators. The first algorithm is based on treating the classical dissipation as a linear amplitude damping-type completely positive trace preserving (CPTP) quantum channel where an unspecified environment interacts with the system of interest and produces the non-unitary evolution. Therefore, the combined system-environment is now closed, and must undergo unitary evolution in the dilated space. The unspecified environment can be modeled by just one ancillary qubit, resulting in an implementation scaling of O(2n−1n2) elementary gates for the total system-environment unitary evolution operator. The second algorithm approximates the non-unitary operators by the Linear Combination of Unitaries (LCU). On exploiting the diagonal structure of the dissipation, we obtain an optimized representation of the non-unitary part, which requires O(2n) elementary gates. Applying the LCU method for a simple dielectric medium with homogeneous dissipation rate, the implementation scaling can be further reduced into O[poly(n)] basic gates. For the particular case of weak dissipation we show that our proposed post-selective dilation algorithms can efficiently delve into the transient evolution dynamics of dissipative systems by calculating the respective implementation circuit depth. A connection of our results with the non-linear-in-normalization-only (NINO) quantum channels is also presented.

Web Service Implementation “AI Yoga Trainer Using Human Pose Estimation” for Web Discovery

This research work is to create a application as web service to make it available on web for web service discovery. The most significant yoga stance is well-known on a global scale and supports the health advantages advocated by the ancient sages. OpenPose estimates that computer vision technology can help yoga practitioners achieve the greatest form and alignment. The integration of the OpenCV, Python, and MediaPipe frameworks is the main emphasis of this research project in order to develop an OpenPose estimation system for yoga poses. A camera is used to capture the yoga practitioner's motions, and a deep learning model is used to predict the important body parts. The algorithm then looks at these essential aspects to determine if the practitioner is doing the pose correctly or whether any adjustments are needed. The device gives the practitioner immediate feedback, allowing them to modify their alignment and posture as necessary. The OpenPose estimate system can help with the practice of yoga by offering detailed visualizations of the essential body parts throughout each posture in addition to providing real-time feedback Practitioners may more clearly understand the right form and alignment and make the required adjustments by utilizing this visualization. Additionally, the OpenPose estimation system enables the tracking of development over time. Practitioners who want to practice alone at home or may not have access to an instructor can also benefit from it. This paper offers a MediaPipe, OpenCV, and Python-based OpenPose estimation system for yoga positions. For optimal form and alignment, the tool may offer real-time feedback and visual representations of key body parts throughout each position. It is useful tool for yoga teachers and students, enhancing yoga's safety and efficacy while supporting the practice.

Optoelectronic properties of fluorine and antimony-doped tin dioxide nanoparticles

Nano-sized tin dioxide SnO2 (TO) has garnered great potential for use in optoelectronic applications, gas sensors, and solar cell technology thanks to its unique properties. These properties can be further enhanced by altering dopants and crystallinity. To achieve these aims, we successfully synthesized nanoparticles of fluorine (F) and antimony (Sb) doped tin dioxide by using a modified sol-gel method. The synthesis process involved supercritical drying of ethyl alcohol. The effect of doping on TO sample microstructure, optical and electrical properties was studied, which is crucial to understanding its potential applications. The X-ray diffraction pattern reveals that the nanoparticles crystallized in the tetragonal rutile structure with a size varying from 22 to 25 nm depending on element doping. The structural analysis was further proved by Raman spectroscopy and FTIR investigations. The optical band gap energy decreases upon the doping, as evidenced by the Eg values of 3.88, 3.87, and 3.72 eV corresponding to, FTO, and ATO respectively. This result proves that our compounds are good candidates for optoelectronic applications. Further, the electrical properties proved the semiconductor behavior for all the samples. Its conductivity is thermally activated, with a decrease in activation energy observed upon the addition of dopants. Moreover, the temperature dependence of the exponents's’of the (ac) conductivity proves that the conduction mechanism varies with dopants. The decrease of s values for undoped and FTO samples revealed a dominant correlated barrier hopping, while the increase of ‘s’ for ATO confirmed the tunneling hopping mechanism. Furthermore, the representation of real and imaginary parts of impedance has also been analyzed. Thus, the obtained results open exciting opportunities for using these samples in many fields that foster the combination of optical and electrical properties.

Enhancing Online Delivery of Practical Labs for Cyber Security Course Using Virtual Laboratories

This research work is to create a application as web service to make it available on web for web service discovery. The most significant yoga stance is well-known on a global scale and supports the health advantages advocated by the ancient sages. OpenPose estimates that computer vision technology can help yoga practitioners achieve the greatest form and alignment. The integration of the OpenCV, Python, and MediaPipe frameworks is the main emphasis of this research project in order to develop an OpenPose estimation system for yoga poses. A camera is used to capture the yoga practitioner's motions, and a deep learning model is used to predict the important body parts. The algorithm then looks at these essential aspects to determine if the practitioner is doing the pose correctly or whether any adjustments are needed. The device gives the practitioner immediate feedback, allowing them to modify their alignment and posture as necessary. The OpenPose estimate system can help with the practice of yoga by offering detailed visualizations of the essential body parts throughout each posture in addition to providing real-time feedback Practitioners may more clearly understand the right form and alignment and make the required adjustments by utilizing this visualization. Additionally, the OpenPose estimation system enables the tracking of development over time. Practitioners who want to practice alone at home or may not have access to an instructor can also benefit from it. This paper offers a MediaPipe, OpenCV, and Python-based OpenPose estimation system for yoga positions. For optimal form and alignment, the tool may offer real-time feedback and visual representations of key body parts throughout each position. It is useful tool for yoga teachers and students, enhancing yoga's safety and efficacy while supporting the practice.

Rethink Motion Information for Occluded Person Re-Identification

Person re-identification aims to identify the same pedestrians captured by various cameras from different viewpoints in multiple scenarios. Occlusion is the toughest problem for practical applications. In video-based ReID tasks, motion information can be easily obtained from sampled frames, and provide discriminative human part representations. However, most motion-based methodologies are designed for video frames which are not suitable for processing single static image input. In this paper, we propose a Motion-Aware Fusion (MAF) network, aiming to acquire motion information from static images in order to improve the performance of ReID tasks. Specifically, a visual adapter is introduced to enable visual feature extraction, either from image or video data. We design a motion consistency task to guide the motion-aware transformer to learn representative human-part motion information and greatly improve the learning quality of features of occluded pedestrians. Extensive experiments on popular holistic, occluded, and video datasets demonstrate the effectiveness of our proposed method. This method outperforms state-of-the-art approaches by improving the mean average precision (mAP) by 1.5% and rank-1 accuracy by 1.2% on the challenging Occluded-REID dataset. At the same time, it surpasses other methods on the MARS dataset with an improvement of 0.2% in mAP and 0.1% in rank-1 accuracy.

Active self-touch restores bodily proprioceptive spatial awareness following disruption by 'rubber hand illusion'.

Bodily self-awareness relies on a constant integration of visual, tactile, proprioceptive, and motor signals. In the 'rubber hand illusion' (RHI), conflicting visuo-tactile stimuli lead to changes in self-awareness. It remains unclear whether other, somatic signals could compensate for the alterations in self-awareness caused by visual information about the body. Here, we used the RHI in combination with robot-mediated self-touch to systematically investigate the role of tactile, proprioceptive and motor signals in maintaining and restoring bodily self-awareness. Participants moved the handle of a leader robot with their right hand and simultaneously received corresponding tactile feedback on their left hand from a follower robot. This self-touch stimulation was performed either before or after the induction of a classical RHI. Across three experiments, active self-touch delivered after-but not before-the RHI, significantly reduced the proprioceptive drift caused by RHI, supporting a restorative role of active self-touch on bodily self-awareness. The effect was not present during involuntary self-touch. Unimodal control conditions confirmed that both tactile and motor components of self-touch were necessary to restore bodily self-awareness. We hypothesize that active self-touch transiently boosts the precision of proprioceptive representation of the touched body part, thus counteracting the visual capture effects that underlie the RHI.

AAformer: Auto-Aligned Transformer for Person Re-Identification.

In person re-identification (re-ID), extracting part-level features from person images has been verified to be crucial to offer fine-grained information. Most of the existing CNN-based methods only locate the human parts coarsely, or rely on pretrained human parsing models and fail in locating the identifiable nonhuman parts (e.g., knapsack). In this article, we introduce an alignment scheme in transformer architecture for the first time and propose the auto-aligned transformer (AAformer) to automatically locate both the human parts and nonhuman ones at patch level. We introduce the "Part tokens (PARTs)", which are learnable vectors, to extract part features in the transformer. A PART only interacts with a local subset of patches in self-attention and learns to be the part representation. To adaptively group the image patches into different subsets, we design the auto-alignment. Auto-alignment employs a fast variant of optimal transport (OT) algorithm to online cluster the patch embeddings into several groups with the PARTs as their prototypes. AAformer integrates the part alignment into the self-attention and the output PARTs can be directly used as part features for retrieval. Extensive experiments validate the effectiveness of PARTs and the superiority of AAformer over various state-of-the-art methods.

ТИПИ ХУДОЖНЬОЇ РЕФЕРЕНЦІЇ: ПРОБЛЕМИ ІНТЕРПРЕТАЦІЇ

The article is devoted to the problems of artistic reference, the ways of its presentation in the artistic text and the possibilities of understanding. The types of artistic reference characteristic of the modern French novel are analyzed. The conducted analysis allows us to talk about three main types – introductory, identifying, undefined. Besides, attention is focused on the interconnection of the categories of reference and modality, as well as on its representation in different parts of architecture and composition.

A differentiated palynological record of the Eemian interglacial in two palaeolakes Niesadna and Parysów from the Garwolin Plain (Central Poland)

Two neighbouring Eemian (MIS 5e) palaeolakes, discovered during works on reambulation of the Garwolin sheet of the Detailed Geological Map of Poland (scale 1: 50 000), revealed differences in the palynological record of the succession of the Eemian interglacial. In the profile of Parysów WH-125, a very pronounced hiatus in the transition from the mesocratic to the telocreatic period, i.e. from the hornbeam to the spruce-fir zone, is visible. In the pollen diagram of Niesadna, the maximum percentages reached by Quercus, Corylus and Carpinus are very low, although this is undoubtedly an Eemian succession. The reasons for this should be sought in the presence of several sedimentary gaps probably caused by the different geological and geomorphological conditions of the Niesadna palaeolake located on a slope as opposed to the Parysów palaeolake, which was formed by the melting of a block of dead ice in the process of deglaciation of the area during the late Saalian (MIS 6). An additional rationale for the hypothesis of extremely unfavourable conditions for undisturbed sedimentation in Niesadna palaeolake is its geological-geomorphological situation that caused admixtures of mineral material in the organic sediment, including sand intercalations, indicating that erosion was taking place. The two developed pollen successions were presented against other published data on the Eemian interglacial in the Garwolin Plain, with particular emphasis on the continuity of the Eemian succession. The palynological results showed that the Parysów WH-125 profile joins several other profiles lacking representation of the upper part of the hornbeam zone and sometimes also the lower part of the spruce-fir zone (e.g. Żabieniec, Jagodne). On the other hand, the presence of distinct sedimentary gaps in the oak, hazel, and hornbeam zone, as is the case in the Niesadna profile, has not been recorded in any of the profiles in the Garwolin Plain examined to date. The Eemian pollen succession in the studied Niesadna and Parysów palaeolakes is presented against the broader background of the course of this interglacial in the European Lowlands.

DES-YOLO: a novel model for real-time detection of casting surface defects.

Surface defect inspection methods have proven effective in addressing casting quality control tasks. However, traditional inspection methods often struggle to achieve high-precision detection of surface defects in castings with similar characteristics and minor scales. The study introduces DES-YOLO, a novel real-time method for detecting castings' surface defects. In the DES-YOLO model, we incorporate the DSC-Darknet backbone network and global attention mechanism (GAM) module to enhance the identification of defect target features. These additions are essential for overcoming the challenge posed by the high similarity among defect characteristics, such as shrinkage holes and slag holes, which can result in decreased detection accuracy. An enhanced pyramid pooling module is also introduced to improve feature representation for small defective parts through multi-layer pooling. We integrate Slim-Neck and SIoU bounding box regression loss functions for real-time detection in actual production scenarios. These functions reduce memory overhead and enable real-time detection of surface defects in castings. Experimental findings demonstrate that the DES-YOLO model achieves a mean average precision (mAP) of 92.6% on the CSD-DET dataset and a single-image inference speed of 3.9 milliseconds. The proposed method proves capable of swiftly and accurately accomplishing real-time detection of surface defects in castings.

Еволюція форм вираження авторського модусу у французькому романі

The article is devoted to the problem of expressing the author's mode in artistic discourse, and specifically to the types of narrators and their change in French literature of the 19th-21st centuries. The main focus is on the internalized narrator as the leading form of the modern French novel, which receives a polymodal status due to the special contamination of the voices of the author, the narrator and the character. The article continues the development of the theory of textual modality, namely, the analysis of linguistic forms of its presentation in the text is made, the main means of expressing the modal meaning and showing the author's attitude are determined. Among them, the subjective epithet, the techniques of speech dislocation, and inserted elements that enhance the emotionality of the work deserve special attention. The evolution of the forms of presentation of the author's point of view, the linguistic personality of the author as a bearer of a certain culture is studied, the features of the new text structure with contaminated types of speech are highlighted. Attention is focused on the interconnectedness of the categories of modality, reference and axiologicality, their representation in different parts of the composition, as well as such categories as modalization and focalization, which are considered as meaning- and text-forming. Special attention is paid to expressive modality, forms of integral polylogue, subjective psychologism of the story. It turns out that a special role begins to be given to the mode of doubt, uncertainty of meaning, mechanisms of reflection, actualization of the communicative and discursive feature of the story, and subjective evaluation. It was found that the perceptive mode characteristic of the beginning of the story creates a split reference, to which the mode of subjective reflection is added. The assimilation of the author's voice to the character's voice leads to the disappearance of bright figurative forms, but the modality of accentuation and clarification appears. Modes of self-reflection, self-ignorance, contradictions to generally accepted stereotypes, the appearance of which was impossible in previous narrative types, deserve special attention. Keywords: modality, author's mode, narrator, subjectivization, interiorized narrative, novel, French language. Information about author: Smushchynska Iryna Viktorivna – doctor of philology, professor, head of the department of theory and practice of translation of Roman languages of Mykola Zerov; Educational and scientific institute of philology; Kyiv national Taras Shevchenko university.

Relational Part-Aware Learning for Complex Composite Object Detection in High-Resolution Remote Sensing Images.

In high-resolution remote sensing images (RSIs), complex composite object detection (e.g., coal-fired power plant detection and harbor detection) is challenging due to multiple discrete parts with variable layouts leading to complex weak inter-relationship and blurred boundaries, instead of a clearly defined single object. To address this issue, this article proposes an end-to-end framework, i.e., relational part-aware network (REPAN), to explore the semantic correlation and extract discriminative features among multiple parts. Specifically, we first design a part region proposal network (P-RPN) to locate discriminative yet subtle regions. With butterfly units (BFUs) embedded, feature-scale confusion problems stemming from aliasing effects can be largely alleviated. Second, a feature relation Transformer (FRT) plumbs the depths of the spatial relationships by part-and-global joint learning, exploring correlations between various parts to enhance significant part representation. Finally, a contextual detector (CD) classifies and detects parts and the whole composite object through multirelation-aware features, where part information guides to locate the whole object. We collect three remote sensing object detection datasets with four categories to evaluate our method. Consistently surpassing the performance of state-of-the-art methods, the results of extensive experiments underscore the effectiveness and superiority of our proposed method.

Learning Mutually Exclusive Part Representations for Fine-Grained Image Classification

Learning Mutually Exclusive Part Representations for Fine-Grained Image Classification

Gait recognition based on multi-feature representation and temporal modeling of periodic parts

Despite the ability of 3D convolutional methods to extract spatio-temporal information simultaneously, they also increase parameter redundancy and computational and storage costs. Previous work that has utilized the 2D convolution method has approached the problem in one of two ways: either using the entire body sequence as input to extract global features or dividing the body sequence into several parts to extract local features. However, global information tends to overlook detailed information specific to each body part, while local information fails to capture relationships between local regions. Therefore, this study proposes a new framework for constructing spatio-temporal representations, which involves extracting and fusing features in a novel manner. To achieve this, we introduce the multi-feature extraction-fusion (MFEF) module, which includes two branches: each branch extracts global features or local features individually, after which they are fused using multiple strategies. Additionally, as gait is a periodic action and different body parts contribute unequally to recognition during each cycle, we propose the periodic temporal feature modeling (PTFM) module, which extracts temporal features from adjacent frame parts during the complete gait cycle, based on the fused features. Furthermore, to capture fine-grained information specific to each body part, our framework utilizes multiple parallel PTFMs to correspond with each body part. We conducted a comprehensive experimental study on the widely used public dataset CASIA-B. Results indicate that the proposed approach achieved an average rank-1 accuracy of 97.2% in normal walking conditions, 92.3% while carrying a bag during walking, and 80.5% while wearing a jacket during walking.

Teknik Pemeriksaan Magnetic Resonance Imaging (MRI) Brain Dengan Penambahan Sequence Axial 3D-Fiesta Pada Kasus Facial Tic

Magnetic Resonance Imaging is a medical device that combines computer technology, strong magnetic fields and radio waves to emulate representations of human body parts and produce more detailed and clear images, one of which is Facial Tic on the Brain. The MRI Brain examination aims to see the anatomy and abnormalities in the brain. This study aims to determine the MRI Brain examination procedure and the role of the Axial 3D Fiesta sequence in cases of Facial Tic. Use of the 3D Fast Imaging Employing Steady State Acquisition (FIESTA) sequence for evaluation of the head nerves. The research method used is descriptive research with a case study approach. Data collection was carried out from July 2023 to August 2023 using a GE 1.5 Tesla MRI aircraft. The subject consisted of a patient with clinical Facial Tic. Data collection was carried out through observation, interviews and documentation. The research results were obtained according to theory using sagittal T1, Coronal PD/T2, Axial PD/T2/T1/FLAIR/EPI, Axial 3D incoherent GRE T1, Axial/oblique SE MT, Axial DWI, Diffusion Tensor Imaging (DTI), Axial perfusion sequences. imaging, while in the field using localizer sequences, Axial DWI, Axial T2 Flair, Axial T2, Axial T1, Axial T2* GRE, Sagittal T1, Coronal T2, and Axial 3D Fiesta

A novel method based on deep reinforcement learning for machining process route planning

Efficient and high-quality machining process route planning is crucial in the realm of manufacturing. Traditional methods heavily rely on human-computer interaction, which can be inefficient. To enhance the efficiency of machining process route planning, this paper introduces a novel framework based on deep reinforcement learning (DRL), designed to automatically generate machining process routes for designated parts. The framework treats machining process route planning as a Markov decision process, making it amenable to DRL techniques. For effective representations of parts, the framework utilizes directed attributed adjacency graphs, wherein nodes represent machining features and edges represent their relationships. To effectively process the graphs, convolutional graph neural networks are employed as the underlying neural networks in the framework. After training, the framework is able to generate efficient machining process routes while adhering to machining process rules. Experimental studies conducted on a number of aircraft structural parts serve as examples to show the feasibility and effectiveness of the proposed approach. The experimental results underscore the effectiveness of the proposed method in machining processes planning for parts and its ability to overcome limitations in traditional methods. As a result, this study contributes to the enhancement of process planning efficiency within the manufacturing domain, carrying practical implications.