Article
Computer Science, Artificial Intelligence
Bongyeong Koo, Han-Soo Choi, Myungjoo Kang
Summary: In this study, a novel weakly supervised object localization (WSOL) method is proposed, which combines a spatial attention branch and a refinement branch for accurate localization. By enhancing spatial information and considering spatial relationships, the proposed method achieves state-of-the-art performance on the CUB-200-2011 and ILSVRC 2012 datasets. The method also demonstrates efficiency with lightweight trainable parameters.
IMAGE AND VISION COMPUTING
(2023)
Article
Computer Science, Information Systems
Songxiang Yang, Lin Ma, Xuezhi Tan
Summary: This paper proposes an end-to-end weakly supervised class-agnostic image retrieval method based on convolutional neural networks. The method preprocesses and clusters the database images to avoid mismatches caused by background mixing, and shows better performance on multiple datasets.
IEEE TRANSACTIONS ON EMERGING TOPICS IN COMPUTING
(2022)
Article
Computer Science, Information Systems
Hatem Ibrahem, Ahmed Diefy Ahmed Salem, Hyun-Soo Kang
Summary: The COFL method utilizes a high-speed convolutional neural network to combine multi-label classification and regression for weakly supervised localization and object detection, achieving clear object localization without requiring bounding box annotations through a masking technique between CAMs and RAMs.
Article
Computer Science, Artificial Intelligence
Jie Lin, Yu Zhan, Wan-Lei Zhao
Summary: This paper proposes a compact instance level feature representation using two CNN pipelines to localize potential instances and generate distinctive features, considering factors such as sensitivity to unknown categories, distinctiveness to different instances, and capability of localizing an instance in an image. This method is suitable for large-scale image collections and is the first work to build instance level representation based on weakly supervised object detection.
Article
Computer Science, Artificial Intelligence
Lian Xu, Mohammed Bennamoun, Farid Boussaid, Wanli Ouyang, Dan Xu
Summary: Weakly supervised semantic segmentation (WSSS) commonly uses Class Activation Mapping (CAM) to generate pseudo semantic labels, but CAM maps have sparse and poor boundaries, resulting in insufficient segmentation supervision. To improve semantic segmentation, we propose joint learning of semantic segmentation and class-agnostic masks using image-level annotations and off-the-shelf saliency maps. We also introduce a cross-task label refinement mechanism and a normalization method for CAM to generate high-quality pseudo semantic segmentation labels.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Wanchun Sun, Xin Feng, Hui Ma, Jingyao Liu
Summary: This article introduces a weakly supervised object localization method based on transformers, which addresses the limitations of traditional methods by introducing hierarchical comprehension of transformers (HiCT), discriminative-based attention layer (DAL), and spatial aware digging module (SADM), and its effectiveness is proven through experiments.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2023)
Article
Computer Science, Artificial Intelligence
Chen -Lin Zhang, Yin Li, Jianxin Wu
Summary: This paper proposes a weakly supervised foreground learning (WSFL) task, which greatly improves weakly supervised object localization (WSOL) and detection (WSOD) by providing groundtruth foreground masks. A complete WSFL pipeline with low computational cost is also introduced, which generates pseudo boxes, learns foreground masks, and does not require any localization annotations. With the help of foreground masks predicted by the WSFL model, state-of-the-art performance is achieved on CUB dataset with 74.37% correct localization accuracy for WSOL, and on VOC07 dataset with 55.7% mean average precision for WSOD. The WSFL model also demonstrates excellent transferability.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Theory & Methods
Bongyeong Koo, Han-Soo Choi, Myungjoo Kang
Summary: The study proposes a new architecture that utilizes FPN and multi-scale information to simplify WSOL optimization problems and improve localization accuracy. In the proposed model, FPN generates multi-scale high-quality feature maps, which are then fused for classification to enable accurate object localization. Experimental results demonstrate the method's superiority over state-of-the-art techniques on datasets, with the added benefit of simplifying optimization challenges.
MULTIDIMENSIONAL SYSTEMS AND SIGNAL PROCESSING
(2021)
Article
Computer Science, Artificial Intelligence
Byeongjoon Kim, Hyunjung Shim, Jongduk Baek
Summary: In this study, a weakly-supervised denoising framework is proposed to generate paired original and noisier CT images from unpaired CT images using a physics-based noise model. The experimental results demonstrate that the method achieves remarkable performances in diagnostic image quality, even superior to fully-supervised CT denoising in terms of signal detectability.
MEDICAL IMAGE ANALYSIS
(2021)
Article
Computer Science, Artificial Intelligence
Yuan Yao, Fang Wan, Wei Gao, Xingjia Pan, Zhiliang Peng, Qi Tian, Qixiang Ye
Summary: This article introduces a method that incorporates vision transformers into weakly supervised object localization to capture long-range semantic dependencies. The TS-CAM method shows superiority in multicategory object localization.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Information Systems
Gautam Kumar, Prateek Keserwani, Partha Pratim Roy, Debi Prosad Dogra
Summary: The paper introduces a logo detection method utilizing weakly supervised learning with CNN to generate a deep saliency map for logo region detection. The method fine-tunes AlexNet, CaffeNet, and VGGNet deep architectures for classification and achieves a mean average precision of 75.83% on the FlickrLogos-32 dataset, outperforming existing fully supervised methods.
MULTIMEDIA TOOLS AND APPLICATIONS
(2021)
Article
Environmental Sciences
Yang Long, Xiaofang Zhai, Qiao Wan, Xiaowei Tan
Summary: This paper proposes a weakly supervised learning method for object recognition in remote sensing images. By using image-level semantic labels and location points, multiple objects can be recognized without relying on bounding box annotations. The experimental results show that the proposed method outperforms existing methods in localizing aircraft and oiltanks in remote sensing images.
Article
Computer Science, Artificial Intelligence
Wenlong Gao, Ying Chen, Yong Peng
Summary: In this paper, a novel cascade attentive dropout strategy is proposed to alleviate the part domination problem and improve the global context module for weakly supervised object detection.
NEURAL PROCESSING LETTERS
(2023)
Article
Computer Science, Artificial Intelligence
Bum Jun Kim, Gyogwon Koo, Hyeyeon Choi, Sang Woo Kim
Summary: Understanding the inner behaviors of deep neural networks is important and visualization methods are widely used. Existing saliency methods show low visualization accuracy, and this study proposes an improved saliency algorithm. The study introduces conservativeness as a property for avoiding redundancy and deficiency in saliency maps, and proposes a new CAM equation with improved theoretical properties. Additionally, the study addresses the problematic practice of using bilinear upsampling and proposes Gaussian upsampling as an improved method. The proposed Extended-CAM method achieves more accurate visualizations compared to existing methods.
COMPUTER VISION AND IMAGE UNDERSTANDING
(2023)
Article
Nuclear Science & Technology
Yu Wang, Qingxu Yao, Quanhu Zhang, He Zhang, Yunfeng Lu, Qimeng Fan, Nan Jiang, Wangtao Yu
Summary: This study proposes an explainable radionuclide identification algorithm based on the convolutional neural network and class activation mapping, which can accurately identify nuclides and explain the identification principles under low count and low signal-to-noise ratio conditions.
NUCLEAR ENGINEERING AND TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Hao Yang, Min Wang, Zhengfei Yu, Hang Zhang, Jinshen Jiang, Yun Zhou
Summary: In this paper, a novel method called CSTTA is proposed for test time adaptation (TTA), which utilizes confidence-based optimization and sample reweighting to better utilize sample information. Extensive experiments demonstrate the effectiveness of the proposed method.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Jin Liu, Ju-Sheng Mi, Dong-Yun Niu
Summary: This article focuses on a novel method for generating a canonical basis for decision implications based on object-induced operators (OE operators). The logic of decision implication based on OE operators is described, and a method for obtaining the canonical basis for decision implications is given. The completeness, nonredundancy, and optimality of the canonical basis are proven. Additionally, a method for generating true premises based on OE operators is proposed.
KNOWLEDGE-BASED SYSTEMS
(2024)
Review
Computer Science, Artificial Intelligence
Kun Bu, Yuanchao Liu, Xiaolong Ju
Summary: This paper discusses the importance of sentiment analysis and pre-trained models in natural language processing, and explores the application of prompt learning. The research shows that prompt learning is more suitable for sentiment analysis tasks and can achieve good performance.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Xiangjun Cai, Dagang Li
Summary: This paper presents a new decomposition mechanism based on learned decomposition mapping. By using a neural network to learn the relationship between original time series and decomposed results, the repetitive computation overhead during rolling decomposition is relieved. Additionally, extended mapping and partial decomposition methods are proposed to alleviate boundary effects on prediction performance. Comparative studies demonstrate that the proposed method outperforms existing RDEMs in terms of operation speed and prediction accuracy.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Xu Wu, Yang Liu, Jie Tian, Yuanpeng Li
Summary: This paper proposes a blockchain-based privacy-preserving trust management architecture, which adopts federated learning to train task-specific trust models and utilizes differential privacy to protect device privacy. In addition, a game theory-based incentive mechanism and a parallel consensus protocol are proposed to improve the accuracy of trust computing and the efficiency of consensus.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Zaiyang Yu, Prayag Tiwari, Luyang Hou, Lusi Li, Weijun Li, Limin Jiang, Xin Ning
Summary: This study introduces a 3D view-based approach that effectively handles occlusions and leverages the geometric information of 3D objects. The proposed method achieves state-of-the-art results on occluded ReID tasks and exhibits competitive performance on holistic ReID tasks.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Yongliang Shi, Runyi Yang, Zirui Wu, Pengfei Li, Caiyun Liu, Hao Zhao, Guyue Zhou
Summary: Neural implicit representations have gained attention due to their expressive, continuous, and compact properties. However, there is still a lack of research on city-scale continual implicit dense mapping based on sparse LiDAR input. In this study, a city-scale continual neural mapping system with a panoptic representation is developed, incorporating environment-level and instance-level modeling. A tailored three-layer sampling strategy and category-specific prior are proposed to address the challenges of representing geometric information in city-scale space and achieving high fidelity mapping of instances under incomplete observation.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Ruihan Hu, Zhi-Ri Tang, Rui Yang, Zhongjie Wang
Summary: Mesh data is crucial for 3D computer vision applications worldwide, but traditional deep learning frameworks have struggled with handling meshes. This paper proposes MDSSN, a simple mesh computation framework that models triangle meshes and represents their shape using face-based and edge-based Riemannian graphs. The framework incorporates end-to-end operators inspired by traditional deep learning frameworks, and includes dedicated modules for addressing challenges in mesh classification and segmentation tasks. Experimental results demonstrate that MDSSN outperforms other state-of-the-art approaches.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Buliao Huang, Yunhui Zhu, Muhammad Usman, Huanhuan Chen
Summary: This paper proposes a novel semi-supervised conditional normalizing flow (SSCFlow) algorithm that combines unsupervised imputation and supervised classification. By estimating the conditional distribution of incomplete instances, SSCFlow facilitates imputation and classification simultaneously, addressing the issue of separated tasks ignoring data distribution and label information in traditional methods.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Deeksha Varshney, Asif Ekbal, Erik Cambria
Summary: This paper focuses on the neural-based interactive dialogue system that aims to engage and retain humans in long-lasting conversations. It proposes a new neural generative model that combines step-wise co-attention, self-attention-based transformer network, and an emotion classifier to control emotion and knowledge transfer during response generation. The results from quantitative, qualitative, and human evaluation show that the proposed models can generate natural and coherent sentences, capturing essential facts with significant improvement over emotional content.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Junchen Ye, Weimiao Li, Zhixin Zhang, Tongyu Zhu, Leilei Sun, Bowen Du
Summary: Modeling multivariate time series has long been a topic of interest for scholars in various fields. This paper introduces MvTS, an open library based on Pytorch, which provides a unified framework for implementing and evaluating these models. Extensive experiments on public datasets demonstrate the effectiveness and universality of the models reproduced by MvTS.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Reham R. Mostafa, Ahmed M. Khedr, Zaher Al Aghbari, Imad Afyouni, Ibrahim Kamel, Naveed Ahmed
Summary: Feature selection is crucial in classification procedures, but it faces challenges in high-dimensional datasets. To overcome these challenges, this study proposes an Adaptive Hybrid-Mutated Differential Evolution method that incorporates the mechanics of the Spider Wasp Optimization algorithm and the concept of Enhanced Solution Quality. Experimental results demonstrate the effectiveness of the method in terms of accuracy and convergence speed, and it outperforms contemporary cutting-edge algorithms.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Ti Xiang, Pin Lv, Liguo Sun, Yipu Yang, Jiuwu Hao
Summary: This paper introduces a Track Classification Model (TCM) based on marine radar, which can effectively recognize and classify shipping tracks. By using a feature extraction network with multi-feature fusion and a dataset production method to address missing labels, the classification accuracy is improved, resulting in successful engineering application in real scenarios.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Zhihao Zhang, Yuan Zuo, Chenghua Lin, Junjie Wu
Summary: This paper proposes a novel unsupervised context-aware quality phrase mining framework called LMPhrase, which is built upon large pre-trained language models. The framework mines quality phrases as silver labels using a parameter-free probing technique on the pre-trained language model BERT, and formalizes the phrase tagging task as a sequence generation problem by fine-tuning on the Sequence to-Sequence pre-trained language model BART. The results of extensive experiments show that LMPhrase consistently outperforms existing competitors in two different granularity phrase mining tasks.
KNOWLEDGE-BASED SYSTEMS
(2024)
Article
Computer Science, Artificial Intelligence
Kemal Buyukkaya, M. Ozan Karsavuran, Cevdet Aykanat
Summary: The study aims to investigate the hybrid parallelization of the Stochastic Gradient Descent (SGD) algorithm for solving the matrix completion problem on a high-performance computing platform. A hybrid parallel decentralized SGD framework with asynchronous inter-process communication and a novel flexible partitioning scheme is proposed to achieve scalability up to hundreds of processors. Experimental results on real-world benchmark datasets show that the proposed algorithm achieves 6x higher throughput on sparse datasets compared to the state-of-the-art, while achieving comparable throughput on relatively dense datasets.
KNOWLEDGE-BASED SYSTEMS
(2024)