Generation Module Research Articles

Cross-Domain Person Re-Identification Based on Multi-Branch Pose-Guided Occlusion Generation

Aiming at the problems caused by a lack of feature matching due to occlusion and fixed model parameters in cross-domain person re-identification, a method based on multi-branch pose-guided occlusion generation is proposed. This method can effectively improve the accuracy of person matching and enable identity matching even when pedestrian features are misaligned. Firstly, a novel pose-guided occlusion generation module is designed to enhance the model’s ability to extract discriminative features from non-occluded areas. Occlusion data are generated to simulate occluded person images. This improves the model’s learning ability and addresses the issue of misidentifying occlusion samples. Secondly, a multi-branch feature fusion structure is constructed. By fusing different feature information from the global and occlusion branches, the diversity of features is enriched. This enrichment improves the model’s generalization. Finally, a dynamic convolution kernel is constructed to calculate the similarity between images. This approach achieves effective point-to-point matching and resolves the problem of fixed model parameters. Experimental results indicate that, compared to current mainstream algorithms, this method shows significant advantages in the first hit rate (Rank-1), mean average precision (mAP), and generalization performance. In the MSMT17→DukeMTMC-reID dataset, after re-ranking (Rerank) and time-tift (Tlift) for the two indicators on Market1501, the mAP and Rank-1 reached 80.5%, 84.3%, 81.9%, and 93.1%. Additionally, the algorithm achieved 51.6% and 41.3% on DukeMTMC-reID→Occluded-Duke, demonstrating good recognition performance on the occlusion dataset.

Self-supervised learning fabric defect segmentation using anomaly generation

PurposeThe problem of fabric defect detection is a particularly challenging task, as the fabric defects occupy only a small portion of the image pixels and it is difficult to collect sufficient samples for training deep learning-based models. The purpose of this work is to present a novel self-supervised learning method to address this problem.Design/methodology/approachIn order to solve the problem of lack of defect samples, based on the fabric-specific degree of texture regularity, we propose an anomaly generation method to create synthetic fabric defects by destroying the normal fabric texture. To improve the detection of defects of different sizes, a global–local parallel detection mechanism is proposed. A self-supervised model including an anomaly generation module, a reconstruction subnetwork and a discriminative subnetwork is established to achieve model training without prior anomaly information.FindingsThe proposed method features self-supervised training, does not require no labelled anomaly data and detects anomalies by distinguishing their distance from normal samples at both global and local levels. When tested on four fabric datasets, our approach outperforms state-of-the-art unsupervised and self-supervised methods and achieves significantly higher localization accuracy.Originality/valueA high-fidelity fabric defects synthesis method is presented to alleviate the problem of collecting numerous fabric defects, providing a reference for other surface anomaly detection problems. The proposed global–local parallel detection mechanism can serve as a reference for other methods dedicated to detecting particular anomalies. The proposed self-supervised network model can effectively locate fabric anomalies without prior labelling information, which could be used as a framework for other model designs.

Model-Driven Deep Neural Network for Enhancing Direction Finding with Commodity 5G gNodeB

Pervasive and high-accuracy positioning has become increasingly important as a fundamental enabler for intelligent connected devices in mobile networks. Nevertheless, current wireless networks heavily rely on pure model-driven techniques to achieve positioning functionality, often succumbing to performance deterioration due to hardware impairments in practical scenarios. Here we reformulate the direction finding or angle-of-arrival (AoA) estimation problem as an image recovery task of the spatial spectrum and propose a new model-driven deep neural network (MoD-DNN) framework. The proposed MoD-DNN scheme comprises three modules: a multi-task autoencoder-based beamformer, a coarray spectrum generation module, and a model-driven deep learning-based spatial spectrum reconstruction module. Our technique enables automatic calibration of angular-dependent phase error thereby enhancing the resilience of direction-finding precision against realistic system non-idealities. We validate the proposed scheme both using numerical simulations and field tests. The results show that the proposed MoD-DNN framework enables effective spectrum calibration and accurate AoA estimation. To the best of our knowledge, this study marks the first successful demonstration of hybrid data-and-model-driven direction finding utilizing readily available commodity 5G gNodeB.

High-Quality Text-to-Image Generation Using High-Detail Feature-Preserving Network

Multistage text-to-image generation algorithms have shown remarkable success. However, the images produced often lack detail and suffer from feature loss. This is because these methods mainly focus on extracting features from images and text, using only conventional residual blocks for post-extraction feature processing. This results in the loss of features, greatly reducing the quality of the generated images and necessitating more resources for feature calculation, which will severely limit the use and application of optical devices such as cameras and smartphones. To address these issues, the novel High-Detail Feature-Preserving Network (HDFpNet) is proposed to effectively generate high-quality, near-realistic images from text descriptions. The initial text-to-image generation (iT2IG) module is used to generate initial feature maps to avoid feature loss. Next, the fast excitation-and-squeeze feature extraction (FESFE) module is proposed to recursively generate high-detail and feature-preserving images with lower computational costs through three steps: channel excitation (CE), fast feature extraction (FFE), and channel squeeze (CS). Finally, the channel attention (CA) mechanism further enriches the feature details. Compared with the state of the art, experimental results obtained on the CUB-Bird and MS-COCO datasets demonstrate that the proposed HDFpNet achieves better performance and visual presentation, especially regarding high-detail images and feature preservation.

STIDNet: Spatiotemporally Integrated Detection Network for Infrared Dim and Small Targets

Infrared dim and small target detection (IRDSTD) aims to obtain target position information from the background, clutter, and noise. However, for infrared dim and small targets with low signal-to-clutter ratios (SCRs), the detection difficulty lies in the fact that their poor local spatial saliency will lead to missed detections and false alarms. In this work, a spatiotemporally integrated detection network (STIDNet) is proposed for IRDSTD. In the network, a spatial saliency feature generation module (SSFGM) employs a U-shaped network to extract deep features from the spatial dimension of the input image in a frame-by-frame manner and splices them based on the temporal dimension to obtain an airtime feature tensor. IRDSTs with direction-of-motion consistency and strong interframe correlation are reinforced, and randomly generated spurious waves, noise, and other false alarms are inhibited via a fixed-weight multiscale motion feature-based 3D convolution kernel (FWMFCK-3D). A mapping from the features to the target probability likelihood map is constructed in a spatiotemporal feature fusion module (STFFM) by performing 3D convolutional fusion on the spatially localized saliency and time-domain motion features. Finally, several ablation and comparison experiments indicate the excellent performance of the proposed network. For infrared dim and small targets with SCRs < 3, the average AUC value still reached 0.99786.

MKER: multi-modal knowledge extraction and reasoning for future event prediction

Humans can predict what will happen shortly, which is essential for survival, but machines cannot. To equip machines with the ability, we introduce the innovative multi-modal knowledge extraction and reasoning (MKER) framework. This framework combines external commonsense knowledge, internal visual relation knowledge, and basic information to make inference. This framework is built on an encoder-decoder structure with three essential components: a visual language reasoning module, an adaptive cross-modality feature fusion module, and a future event description generation module. The visual language reasoning module extracts the object relationships among the most informative objects and the dynamic evolution of the relationship, which comes from the sequence scene graphs and commonsense graphs. The long short-term memory model is employed to explore changes in the object relationships at different times to form a dynamic object relationship. Furthermore, the adaptive cross-modality feature fusion module aligns video and language information by using object relationship knowledge as guidance to learn vision-language representation. Finally, the future event description generation module decodes the fused information and generates the language description of the next event. Experimental results demonstrate that MKER outperforms existing methods. Ablation studies further illustrate the effectiveness of the designed module. This work advances the field by providing a way to predict future events, enhance machine understanding, and interact with dynamic environments.

Efficiently Exploiting Muti-Level User Initial Intent for Session-Based Recommendation

Session-based recommendation (SBR) aims to predict potential user interactions within an anonymous session. It utilizes learned user interests to recommend items. As research has progressed, researchers have shifted towards exploring user initial intent, which can provide practical guidance for item selection. However, a significant limitation of the current methodologies is that they often assume the first item in a session as the initial intent, neglecting the possibility of a random initial click. Additionally, these methods typically merge the initial intent with the session representation without considering dynamic changes in user interests. To address these challenges, we propose an innovative approach named Efficiently Exploiting Muti-level User Initial Intent for (EMUI) for session-based recommendation. This approach includes a multi-level initial-intent generation module (MIGM) and an interest matching module (IMM). Specifically, the MIGM is designed to extract a more comprehensive representation of user initial intent from various levels, effectively mitigating the issue of random initial clicks. Furthermore, we propose the IMM to ensure alignment between dynamic interests and user initial intent. The IMM identifies components within multi-level user initial intent that correlate with dynamic interests, thereby enhancing session representation and, ultimately, improving recommendation performance. In addition, in order to optimize the initial user intent at each level, we introduce a contrastive learning task to maximize the use of user initial intent at each level. A considerable number of experiments on three real-world datasets have shown that our EMUI has significantly enhanced the recommendation accuracy over state-of-the-art methods.

Automatic image generation and stage prediction of breast cancer immunobiological through a proposed IHC-GAN model.

Invasive breast cancer diagnosis and treatment planning require an accurate assessment of human epidermal growth factor receptor 2 (HER2) expression levels. While immunohistochemical techniques (IHC) are the gold standard for HER2 evaluation, their implementation can be resource-intensive and costly. To reduce these obstacles and expedite the procedure, we present an efficient deep-learning model that generates high-quality IHC-stained images directly from Hematoxylin and Eosin (H&E) stained images. We propose a new IHC-GAN that enhances the Pix2PixHD model into a dual generator module, improving its performance and simplifying its structure. Furthermore, to strengthen feature extraction for HE-stained image classification, we integrate MobileNetV3 as the backbone network. The extracted features are then merged with those generated by the generator to improve overall performance. Moreover, the decoder's performance is enhanced by providing the related features from the classified labels by incorporating the adaptive instance normalization technique. The proposed IHC-GAN was trained and validated on a comprehensive dataset comprising 4,870 registered image pairs, encompassing a spectrum of HER2 expression levels. Our findings demonstrate promising results in translating H&E images to IHC-equivalent representations, offering a potential solution to reduce the costs associated with traditional HER2 assessment methods. We extensively validate our model and the current dataset. We compare it with state-of-the-art techniques, achieving high performance using different evaluation metrics, showing 0.0927 FID, 22.87 PSNR, and 0.3735 SSIM. The proposed approach exhibits significant enhancements over current GAN models, including an 88% reduction in Frechet Inception Distance (FID), a 4% enhancement in Learned Perceptual Image Patch Similarity (LPIPS), a 10% increase in Peak Signal-to-Noise Ratio (PSNR), and a 45% reduction in Mean Squared Error (MSE). This advancement holds significant potential for enhancing efficiency, reducing manpower requirements, and facilitating timely treatment decisions in breast cancer care.

End‐to‐End Crystal Structure Prediction from Powder X‐Ray Diffraction

AbstractPowder X‐ray diffraction (PXRD) is a prevalent technique in materials characterization. While the analysis of PXRD often requires extensive human manual intervention, and most automated method only achieved at coarse‐grained level. The more difficult and important task of fine‐grained crystal structure prediction from PXRD remains unaddressed. This study introduces XtalNet, the first equivariant deep generative model for end‐to‐end crystal structure prediction from PXRD. Unlike previous crystal structure prediction methods that rely solely on composition, XtalNet leverages PXRD as an additional condition, eliminating ambiguity and enabling the generation of complex organic structures with up to 400 atoms in the unit cell. XtalNet comprises two modules: a Contrastive PXRD‐Crystal Pretraining (CPCP) module that aligns PXRD space with crystal structure space, and a Conditional Crystal Structure Generation (CCSG) module that generates candidate crystal structures conditioned on PXRD patterns. Evaluation on two MOF datasets (hMOF‐100 and hMOF‐400) demonstrates XtalNet's effectiveness. XtalNet achieves a top‐10 Match Rate of 90.2% and 79% for hMOF‐100 and hMOF‐400 in conditional crystal structure prediction task, respectively. XtalNet enables the direct prediction of crystal structures from experimental measurements, eliminating the need for manual intervention and external databases. This opens up new possibilities for automated crystal structure determination and the accelerated discovery of novel materials.

Evaluating ChatGPT and Google Gemini Performance and Implications in Turkish Dental Education.

Artificial intelligence (AI) has emerged as a transformative tool in education, particularly in specialized fields such as dentistry. This study evaluated the performance of four advanced AI models - ChatGPT-4o (San Francisco, CA: OpenAI), ChatGPT-o1, Gemini 1.5 Pro (Mountain View, CA: Google LLC), and Gemini 2.0 Advanced, in the Turkish Dental Specialty Examination (DUS) for 2020 and 2021. A total of 240 questions, comprising 120 questions per year from basic and clinical sciences, were analyzed. AI models were assessed based on their accuracy in providing correct answers compared to the official answer keys. For the 2020 DUS, ChatGPT-o1 and Gemini 2.0 Advanced achieved the highest accuracy rates of 93.70% and 96.80%, respectively, with net scores of 112.50 and 115 out of 120 questions. ChatGPT-4o and Gemini 1.5 Pro followed with accuracy rates of 83.33% and 85.40%. For the 2021 DUS, ChatGPT-o1 again demonstrated the highest accuracy at 97.88% (115.50 net score), closely followed by Gemini 2.0 Advanced at 96.82% (114.25 net score). Overall, ChatGPT-4o and Gemini 1.5 Pro scored lower for 2021, achieving accuracy rates of 88.35% and 93.64%, respectively. Combining results from both years (238 total questions), ChatGPT-o1 and Gemini 2.0 Advanced achieved accuracy rates of 97.46% (230 correct answers, 95% CI: 94.62%, 100.00%) and 97.90% (231 correct answers, 95% CI: 94.62%, 100.00%), respectively, significantly outperforming ChatGPT-4o (88.66%, 211 correct answers, 95% CI: 85.43%, 91.89%) and Gemini 1.5 Pro (91.60%, 218 correct answers, 95% CI: 87.75%, 95.45%). Statistical analysis revealed significant differences among the models (p = 0.0002). Pairwise comparisons demonstrated that ChatGPT-4o underperformed significantly compared to ChatGPT-o1 (p = 0.0016) and Gemini 2.0 Advanced (p = 0.0007) after Bonferroni correction. The consistently high accuracy rates and narrow confidence intervals for the top-performing models underscore their superior reliability and performance in answering the DUS questions. Generative AI modules such as ChatGPT-01 and Gemini 2.0 have the potential to enhance dental board exam preparation through question evaluation. While the AI modules appear to outperform humans on DUS questions, the study raises a concern about the ethical uses of AI and the true justification and value of DUS examinations as dental competency examinations. A higher level of knowledge evaluation should be considered. This research contributes to the growing body of literature on AI applications in specialized knowledge domains and provides a foundation for further exploration of its integration into dental education.

Template‐based universal adversarial attack for synthetic aperture radar automatic target recognition network

AbstractExisting synthetic aperture radar (SAR) adversarial attack algorithms primarily focus on the digital image domain, and constructing adversarial examples in real‐world scenarios presents significant and challenging hurdles. This study proposes the template‐based universal adversarial attack (TUAA) algorithm. Initially, a SAR interference template generator module is constructed to derive a universal adversarial perturbation in the image domain. The designed loss function guides the parameter updating of the generator, thereby improving the attack effectiveness and perturbation concealment. Subsequently, a SAR jamming signal generator module is developed, which swiftly generates the interference signal using the range convolutional and azimuth multiplication modulation jamming method. Consequently, the victim model can be effectively targeted by merely transmitting the jamming signal to the SAR receiver. Experimental results show that TUAA reduces the recognition rate of four typical DNN models to less than 15% under acceptable time costs and image deformation.

An Adaptive Multimodal Fusion Network Based on Multilinear Gradients for Visual Question Answering

As an interdisciplinary field of natural language processing and computer vision, Visual Question Answering (VQA) has emerged as a prominent research focus in artificial intelligence. The core of the VQA task is to combine natural language understanding and image analysis to infer answers by extracting meaningful features from textual and visual inputs. However, most current models struggle to fully capture the deep semantic relationships between images and text owing to their limited capacity to comprehend feature interactions, which constrains their performance. To address these challenges, this paper proposes an innovative Trilinear Multigranularity and Multimodal Adaptive Fusion algorithm (TriMMF) that is designed to improve the efficiency of multimodal feature extraction and fusion in VQA tasks. Specifically, the TriMMF consists of three key modules: (1) an Answer Generation Module, which generates candidate answers by extracting fused features and leveraging question features to focus on critical regions within the image; (2) a Fine-grained and Coarse-grained Interaction Module, which achieves multimodal interaction between question and image features at different granularities and incorporates implicit answer information to capture complex multimodal correlations; and (3) an Adaptive Weight Fusion Module, which selectively integrates coarse-grained and fine-grained interaction features based on task requirements, thereby enhancing the model’s robustness and generalization capability. Experimental results demonstrate that the proposed TriMMF significantly outperforms existing methods on the VQA v1.0 and VQA v2.0 datasets, achieving state-of-the-art performance in question–answer accuracy. These findings indicate that the TriMMF effectively captures the deep semantic associations between images and text. The proposed approach provides new insights into multimodal interaction and fusion research, combining domain adaptation techniques to address a broader range of cross-domain visual question answering tasks.

Offset attention with seed generation for point cloud completion

Abstract Point cloud data acquired through 3D scanning is frequently subject to fragmentation due to the constraints of the scanner’s field of view and occlusions within the scanned object. The ensuing incompleteness in the data can significantly degrade the accuracy of subsequent computational tasks. Traditional methods for predicting complete point clouds from these fragments often fail to capture the fine-grained local details, leading to inaccurate reconstructions. In this work, we introduce a novel neural network architecture designed for point cloud completion that addresses these limitations.Our network accepts an incomplete point cloud and employs a multi-scale feature extraction module, which integrates an offset attention mechanism alongside a feature aggregation module operating across various scales. This dual approach significantly bolsters the network’s capacity to discern both local and global features inherent in the point cloud data. Furthermore, we incorporate a seed generation module within our missing point cloud generator, harnessing a hierarchical feature pyramid network to forecast the entirety of the point cloud. This innovative strategy allows our network to accurately predict the structure of missing regions.Empirical evaluations conducted on the Shapenet-Part and ModelNet10 datasets substantiate the efficacy of our proposed methodology. Our approach outperforms the state-of-the-art PF-Net algorithm, achieving a remarkable reduction in chamfer distance by 16.15$\%$ and 41.87$\%$ on the respective datasets. Visual inspection of the results underscores the robust generalization capabilities of our algorithm, which is particularly evident in scenarios with limited dataset sizes. It adeptly predicts the contours of the missing regions and synthesizes a more comprehensive point cloud shape.

Visual Story Generation Model Guided by Multi Granularity Image Information

Visual story generation, which involves generating short stories from sequential images, has become a core task at the intersection of computer vision and natural language processing. However, existing methods suffer from a bias in the concept predicates predicted, leading to a semantic gap between the generated stories and the images. This paper proposes a novel visual story generation model that utilizes muliti granularity image information to guide the generation process and correct the bias in concept predicates, resulting in more image-consistent stories. The proposed model consists of two stages: In the first stage, a set of concepts predicates is predicted from the image and enriched with external knowledge, and the most suitable concepts for story generation are selected. In the second stage, fine-grained image information are utilized to integrate image information into the story generation module, improving the bias in concept predicates. The image theme information and the generated results of previous moments are used as prompts to guide the story generation module. Experimental results show that the proposed model outperforms baseline models in all evaluation metrics. Specifically, the BLEU-1, BLEU-2, BLEU-3, and BLEU-4 metrics are improved by 4.0, 3.8, 3.02, and 1.98 percentage points, respectively, and the METEOR metric is improved by 1.4 percentage points. The generated stories are more consistent with the image content, maintain a consistent theme, and enhance coherence between contexts.

Advanced self-sensing road stud: integrating deep learning-based speed detection and energy harvesting

Abstract This study introduces a self-powered and self-sensing vehicle speed detection sensor, representing a significant advancement in transportation. The system employs mechanical components like a slider crank, bevel gears, and one-way bearings for unidirectional rotation, converting translational motion into electrical energy upon the impact of vehicle tyres on road studs. The electrical power generation module, including a DC generator, rectifier, and battery circuit, captures and stores this energy. In addition to energy harvesting, the system integrates a deep learning model using long short-term memory (LSTM) networks to precisely calculate vehicle speed from the displacement signals of the road studs. Displacement data from an ultrasonic distance sensor (SR-04) is processed and fed into the LSTM network, achieving a classification accuracy of 98.90% for vehicle speed categories of low, medium, high, and overspeed. A mathematical model and MATLAB Simscape simulations were developed, followed by experimental validation using a mechanical testing and sensing system under laboratory conditions. Lab-scale testing, a maximum output power of 3.72 W and an efficiency of 62.7% were recorded at 8 Hz. Field tests were performed at various vehicle speeds. A peak voltage output of 10 V was recorded for a single phase of a three-phase DC generator at 15 km h−1. The displacement sensor beneath the road stud was used to record the relative time signal between adjacent peaks to calculate vehicle speed. The sensor is sustainable in energy and easily installable without infrastructure changes, enhances transportation efficiency, and is useful for traffic management, road safety, and smart transportation networks.

Generative Adversarial Network-Based Distortion Reduction Adapted to Peak Signal-to-Noise Ratio Parameters in VVC

In order to address the issues of image quality degradation and distortion that arise in the context of video transmission coding and decoding, a method based on an enhanced version of CycleGAN is put forth. The lightweight attention module is integrated into the residual block of the generator module structure, thereby facilitating the extraction of image details and motion compensation. Furthermore, the perceptual function LPIPS loss is increased to align the image restoration effect more closely with human perception. Additionally, the network training method is modified, and the original image is divided into 128 × 128 small blocks for training, thus enhancing the network’s accuracy in restoring details. The experimental results demonstrate that the algorithm attains an average PSNR value of 30.1147 on the publicly accessible YUV sequence dataset, YUV Trace Dataset, which is a 9.02% enhancement compared to the original network. Additionally, the LPIPS value reaches 0.2639, representing a 10.42% reduction, and effectively addresses the issue of image quality deterioration during transmission.

Emotion Analysis from Face and Speech: A Comprehensive Review of Current Techniques and Models

Emotion detection in facial expressions and speech plays a crucial role in enhancing interactive platforms, particularly in learning and assessment systems. This study explores advanced techniques for integrating dynamic mock test generation and interview simulation modules in an Advanced Placement Preparation Platform. The dynamic mock test uses real-time feedback to recommend questions based on a student’s performance, leveraging machine learning algorithms for adaptive learning. Additionally, the interview simulation module incorporates facial expression recognition using Convolutional Neural Networks (CNNs) and speech analysis using Recurrent Neural Networks (RNNs) to evaluate student performance. Initial results show that traditional models struggle with real-time adaptability and emotion classification accuracy, underscoring the need for specialized algorithms for complex data. To address these limitations, the system evaluates deep learning models designed for adaptive learning and emotion analysis, such as transfer learning models in emotion detection. The findings highlight the potential for using multimodal data to improve user engagement and performance evaluation in educational settings, paving the way for more immersive and intelligent learning platforms.

A comparative study of virtual synchronous generator and sinusoidal pulse width modulation in a wind high power conversion chain

The virtual synchronous generator (VSG) and sinusoidal pulse width modulation (SPWM) are two prominent control strategies that have attracted particular interest recently. In this paper, we compare these two inverter control strategies in a 5MW wind power conversion chain. The studied conversion chain includes a wind turbine, a permanent magnet synchronous generator, the power converters, namely the uncontrolled rectifier, and a two-stage inverter connected to the grid via an LCL filter. Our study of the two control methods shows that both strategies reduce the total harmonic distortion (THD) while respecting the grid connection conditions. The simulation results manifest that the VSG strategy has a better THD reduction of 0.99 % which is improved compared to the SPWM with a THD of 1.33%.

Attribute Prototype-guided Iterative Scene Graph for Explainable Radiology Report Generation.

The potential of automated radiology report generation in alleviating the time-consuming tasks of radiologists is increasingly being recognized in medical practice. Existing report generation methods have evolved from using image-level features to the latest approach of utilizing anatomical regions, significantly enhancing interpretability. However, directly and simplistically using region features for report generation compromises the capability of relation reasoning and overlooks the common attributes potentially shared across regions. To address these limitations, we propose a novel region-based Attribute Prototype-guided Iterative Scene Graph generation framework (AP-ISG) for report generation, utilizing scene graph generation as an auxiliary task to further enhance interpretability and relational reasoning capability. The core components of AP-ISG are the Iterative Scene Graph Generation (ISGG) module and the Attribute Prototype-guided Learning (APL) module. Specifically, ISSG employs an autoregressive scheme for structural edge reasoning and a contextualization mechanism for relational reasoning. APL enhances intra-prototype matching and reduces inter-prototype semantic overlap in the visual space to fully model the potential attribute commonalities among regions. Extensive experiments on the MIMIC-CXR with Chest ImaGenome datasets demonstrate the superiority of AP-ISG across multiple metrics.

Pixel-Centric Context Perception Network for Camouflaged Object Detection.

Camouflaged object detection (COD) aims to identify object pixels visually embedded in the background environment. Existing deep learning methods fail to utilize the context information around different pixels adequately and efficiently. In order to solve this problem, a novel pixel-centric context perception network (PCPNet) is proposed, the core of which is to customize the personalized context of each pixel based on the automatic estimation of its surroundings. Specifically, PCPNet first employs an elegant encoder equipped with the designed vital component generation (VCG) module to obtain a set of compact features rich in low-level spatial and high-level semantic information across multiple subspaces. Then, we present a parameter-free pixel importance estimation (PIE) function based on multiwindow information fusion. Object pixels with complex backgrounds will be assigned with higher PIE values. Subsequently, PIE is utilized to regularize the optimization loss. In this way, the network can pay more attention to those pixels with higher PIE values in the decoding stage. Finally, a local continuity refinement module (LCRM) is used to refine the detection results. Extensive experiments on four COD benchmarks, five salient object detection (SOD) benchmarks, and five polyp segmentation benchmarks demonstrate the superiority of PCPNet with respect to other state-of-the-art methods.