检索结果-内蒙古大学图书馆

DINOv2-powered Few-Shot Semantic Segmentation: A Unified Framework via Cross-Model Distillation and 4D Correlation Mining

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Zhuo, Wei Tang, Zhiyue Xue, Wufeng Ding, Hao Shen, Linlin School of Artificial Intelligence Shenzhen University Shenzhen518060 China National Engineering Laboratory of Big Data System Computing Technology Shenzhen University China School of Biomedical Engineering Shenzhen University Medical School Shenzhen University China School of Computer Science and Software Engineering Shenzhen University China

Few-shot semantic segmentation has gained increasing interest due to its generalization capability, i.e., segmenting pixels of novel classes requiring only a few annotated images. Prior work has focused on meta-learning for support-query matching, with extensive development in both prototype-based and aggregation-based methods. To address data scarcity, recent approaches have turned to foundation models to enhance representation transferability for novel class segmentation. Among them, a hybrid dual-modal framework including both DINOv2 and SAM has garnered attention due to their complementary capabilities. We wonder "can we build a unified model with knowledge from both foundation models?" To this end, we propose FS-DINO, with only DINOv2’s encoder and a lightweight segmenter. The segmenter features a bottleneck adapter, a meta-visual prompt generator based on dense similarities and semantic embeddings, and a decoder. Through coarse-to-fine cross-model distillation, we effectively integrate SAM’s knowledge into our lightweight segmenter, which can be further enhanced by 4D correlation mining on support-query pairs. Extensive experiments on COCO-20i, PASCAL-5i, and FSS-1000 demonstrate the effectiveness and superiority of our method. Copyright © 2025, The Authors. All rights reserved.

关键词： Semantic Segmentation

Hybrid attention network based on progressive embedding scale-context for crowd counting

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Wang, Fusen Sang, Jun Wu, Zhongyuan Liu, Qi Sang, Nong Key Laboratory of Dependable Service Computing in Cyber Physical Society of Ministry of Education Chongqing University Chongqing400044 China School of Big Data and Software Engineering Chongqing University Chongqing401331 China School of Artificial Intelligence and Automation Huazhong University of Science and Technology Wuhan430000 China

The existing crowd counting methods usually adopted attention mechanism to tackle background noise, or applied multi-level features or multi-scales context fusion to tackle scale variation. However, these ap- proaches deal with these two problems separately. In this paper, we propose an Hybrid Attention Network (HAN) by employing Progressive Embedding Scale-context (PES) information, which enables the network to simultaneously suppress noise and adapt head scale variation. We build the hybrid attention mechanism through paralleling spatial attention and channel attention module, which makes the network to focus more on the human head area and reduce the interference of background objects. Besides, we embed certain scale-context to the hybrid attention along the spatial and channel dimensions for alleviating these count- ing errors caused by the variation of perspective and head scale. Finally, we propose a progressive learning strategy through cascading multiple hybrid attention modules with embedding diαerent scale-context, which can gradually integrate diαerent scale-context information into the current feature map from global to local. Ablation experiments provides that the network architecture can gradually learn multi-scale features and suppress background noise. Extensive experiments demonstrate that HANet obtain state-of-the-art counting performance on four mainstream datasets. Copyright © 2021, The Authors. All rights reserved.

关键词： Embeddings

Learning referee evaluation and assessing action quality from coarse to fine in diving sport

学校读者我要写书评

暂无评论

Neurocomputing 2025年

作者： Hong-Ming Qiu Hong-Bo Zhang Qing Lei Jing-Hua Liu Ji-Xiang Du Department of Computer Science and Technology Huaqiao University Xiamen 361021 China Fujian Key Laboratory of Big Data Intelligence and Security Huaqiao University Xiamen 361021 China Key Laboratory of Computer Vision and Machine Learning (Huaqiao University) Fujian Province University Xiamen 361021 China Xiamen Key Laboratory of Computer Vision and Pattern Recognition Huaqiao University Xiamen 361021 China

Intelligently assessing the quality of athletic performances in sports scenarios remains a fascinating challenge in computer vision. However, unraveling the subtle distinctions between two similar actions in videos and mapping those video representations to quality scores remain significant obstacles. To address these challenges, this work redefines the paradigm of quality score estimation from traditional relative quality score prediction to relative referee score prediction. To make this shift, a cross-feature fusion module rooted in Transformer-based video representation is introduced, to improve pairwise video feature learning in the realm of action quality assessment. Then, a novel contrastive action parsing decoder module generates mid-level representations to effectively connect visual features with detailed quality scores. Both modules utilize cross-attention mechanisms; the former refines the pairwise video features to represent the differences between video pairs, while the latter updates the input queries corresponding to each referee’s evaluation. Finally, to achieve precise quality score estimation, we introduce a meticulous coarse-to-fine decision process, integrating a score classifier and offset regressor. After validation on challenging diving datasets, including MTL-AQA, FineDiving, and TASD-2, the experimental results show that the proposed approach demonstrates effectiveness and feasibility when compared with state-of-the-art methods.

关键词：

A Survey of Knowledge Tracing: Models, Variants, and Applications

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Shen, Shuanghong Liu, Qi Huang, Zhenya Zheng, Yonghe Yin, Minghao Wang, Minjuan Chen, Enhong The School of Data Science University of Science and Technology of China State Key Laboratory of Cognitive Intelligence Anhui Hefei230026 China The School of Computer Science and Techonology University of Science and Technology of China State Key Laboratory of Cognitive Intelligence Institute of Artificial Intelligence Hefei Comprehensive National Science Center Anhui Hefei230026 China The Research Institute of Science Education Faculty of Education Beijing Normal University Haidian Beijing100875 China The School of Information Science and Technology Northeast Normal University Changchun Jilin130117 China The Education University of Hong Kong 999077 Hong Kong San Diego State University San DiegoCA92182 United States

Modern online education has the capacity to provide intelligent educational services by automatically analyzing substantial amounts of student behavioral data. Knowledge Tracing (KT) is one of the fundamental tasks for student behavioral data analysis, aiming to monitor students' evolving knowledge state during their problem-solving process. In recent years, a substantial number of studies have concentrated on this rapidly growing field, significantly contributing to its advancements. In this survey, we will conduct a thorough investigation of these progressions. Firstly, we present three types of fundamental KT models with distinct technical routes. Subsequently, we review extensive variants of the fundamental KT models that consider more stringent learning assumptions. Moreover, the development of KT cannot be separated from its applications, thereby we present typical KT applications in various scenarios. To facilitate the work of researchers and practitioners in this field, we have developed two open-source algorithm libraries: Edudata that enables the download and preprocessing of KT-related datasets, and EduKTM that provides an extensible and unified implementation of existing mainstream KT models. Finally, we discuss potential directions for future research in this rapidly growing field. We hope that the current survey will assist both researchers and practitioners in fostering the development of KT, thereby benefiting a broader range of students. Copyright © 2021, The Authors. All rights reserved.

关键词： Students

Deep Reinforcement Learning with Transformers for Text Adventure Games

学校读者我要写书评

暂无评论

Deep Reinforcement Learning with Transformers for Text Adven...

IEEE Symposium on Computational intelligence and Games, CIG

作者： Yunqiu Xu Ling Chen Meng Fang Yang Wang Chengqi Zhang Centre for Artificial Intelligence University of Technology Sydney Sydney Australia Tencent Robotics X Key Laboratory of Knowledge Engineering with Big Data (Ministry of Education) Hefei University of Technology China

ISBN: (数字)9781728145334

ISBN: (纸本)9781728145341

In this paper, we study transformers for text-based games. As a promising replacement of recurrent modules in Natural Language Processing (NLP) tasks, the transformer architecture could be treated as a powerful state representation generator for reinforcement learning. However, the vanilla transformer is neither effective nor efficient to learn with a huge amount of weight parameters. Unlike existing research that encodes states using LSTMs or GRUs, we develop a novel lightweight transformer-based representation generator featured with reordered layer normalization, weight sharing and block-wise aggregation. The experimental results show that our proposed model not only solves single games with much fewer interactions, but also achieves better generalization on a set of unseen games. Furthermore, our model outperforms state-of-the-art agents in a variety of man-made games.

关键词： Games Generators Task analysis Reinforcement learning Logic gates Buildings Computer architecture

Multi-stage image denoising with the wavelet transform

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Tian, Chunwei Zheng, Menghua Zuo, Wangmeng Zhang, Bob Zhang, Yanning Zhang, David School of Software Northwestern Polytechnical University Shaanxi Xi’an710129 China National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology Shaanxi Xi’an710129 China School of Computer Science and Technology Harbin Institute of Technology Heilongjiang Harbin150001 China Peng Cheng Laboratory Guangdong Shenzhen518055 China Department of Computer and Information Science University of Macau 999078 China School of Computer Science Northwestern Polytechnical University Shaanxi Xi’an710129 China Guangdong Shenzhen518172 China Shenzhen Institute of Artificial Intelligence and Robotics for Society Shenzhen China

Deep convolutional neural networks (CNNs) are used for image denoising via automatically mining accurate structure information. However, most of existing CNNs depend on enlarging depth of designed networks to obtain better denoising performance, which may cause training difficulty. In this paper, we propose a multi-stage image denoising CNN with the wavelet transform (MWDCNN) via three stages, i.e., a dynamic convolutional block (DCB), two cascaded wavelet transform and enhancement blocks (WEBs) and a residual block (RB). DCB uses a dynamic convolution to dynamically adjust parameters of several convolutions for making a tradeoff between denoising performance and computational costs. WEB uses a combination of signal processing technique (i.e., wavelet transformation) and discriminative learning to suppress noise for recovering more detailed information in image denoising. To further remove redundant features, RB is used to refine obtained features for improving denoising effects and reconstruct clean images via improved residual dense architectures. Experimental results show that the proposed MWDCNN outperforms some popular denoising methods in terms of quantitative and qualitative analysis. Codes are available at https://***/hellloxiaotian/MWDCNN. © 2022, CC BY-NC-ND.

关键词： Wavelet transforms

Model-Based Reinforcement Learning for Quantized Federated Learning Performance Optimization

学校读者我要写书评

暂无评论

Model-Based Reinforcement Learning for Quantized Federated L...

GLOBECOM 2022 - 2022 IEEE Global Communications Conference

作者： Nuocheng Yang Sihua Wang Mingzhe Chen Christopher G. Brinton Changchuan Yin Walid Saad Shuguang Cui Beijing Laboratory of Advanced Information Network Beijing University of Posts and Telecommunications Beijing China State Key Laboratory Of Networking And Switching Technology Beijing University of Posts and Telecommunications Beijing China Department of Electrical and Computer Engineering Institute for Data Science and Computing University of Miami Coral Gables FL USA School of Electrical and Computer Engineering Purdue University West Lafayette IN USA Bradley Department of Electrical and Computer Engineering Virginia Tech Arlington VA USA Shenzhen Research Institute of Big Data (SRIBD) and the Future Network of Intelligence Institute (FNii) Chinese University of Hong Kong Shenzhen China

ISBN: (纸本)9781665435413

This paper considers improving wireless communication and computation efficiency in federated learning (FL) via model quantization. In the proposed bitwidth FL scheme, edge devices train and transmit quantized versions of their local FL model parameters to a coordinating server, which, in turn, aggregates them into a quantized global model and synchronizes the devices. With the goal of jointly determining the set of participating devices in each training iteration and the bitwidths employed at the devices, we pose an optimization problem for minimizing the training loss of quantized FL under a device sampling budget and delay requirement. Our analytical results show that the improvement of FL training loss between two consecutive iterations depends on not only the device selection and quantization scheme, but also on several parameters inherent to the model being learned. As a result, we propose, a model-based reinforcement learning (RL) method to optimize action selection over iterations. Compared to model-free RL, the proposed approach leverages the derived mathematical characterization of the FL training process to discover an effective device selection and quantization scheme without imposing additional device communication overhead. Numerical evaluations show that the proposed FL framework can achieve the same classification performance while reducing the number of training iterations needed for convergence by 20% compared to model-free RL-based FL.

关键词： Training Performance evaluation Wireless communication Quantization (signal) Computational modeling Reinforcement learning Mathematical models

Classification of transgenic corn varieties using terahertz spectroscopy and convolutional neural network

学校读者我要写书评

暂无评论

Journal of Food Composition and Analysis 2025年 145卷

作者： Jiang, Yuying Wen, Xixi Ge, Hongyi Li, Guangming Chen, Hao Jiang, Mengdie Sun, Qingcheng Wei, Shilei Li, Peng Key Laboratory of Grain Information Processing and Control Henan University of Technology Ministry of Education Zhengzhou 450001 China Henan Provincial Key Laboratory of Grain Photoelectric Detection and Control Henan University of Technology Zhengzhou 450001 China School of Artificial Intelligence and Big Data Henan University of Technology Zhengzhou 450001 China School of Information Science and Engineering Henan University of Technology Zhengzhou 450001 China Institute for Complexity Science Henan University of Technology Zhengzhou 450001 China

Corn is the largest grain crop in China. With the increasing use of transgenic corn in agricultural production, accurately identifying transgenic corn varieties has become crucial for ensuring food safety and quality. This study combines terahertz time-domain spectroscopy (THz-TDS) technology with a convolutional neural network (CNN) to classify five transgenic corn varieties. First, the time-domain spectra of transgenic corn were collected, and the frequency-domain spectra and absorption coefficient spectra were calculated. Next, abnormal data were removed using the isolation forest algorithm. The data were then preprocessed using the first derivative (FD) and principal component analysis (PCA) to extract effective spectral features. Based on these features, classification models were constructed using support vector machine (PCA-SVM), grid search optimized SVM (PCA-GS-SVM), particle swarm optimization SVM (PCA-PSO-SVM), and hippopotamus optimization SVM (PCA-HO-SVM). Additionally, a one-dimensional CNN (1D-CNN) model was built using the full-band spectral data. The results indicate that the performance of the 1D-CNN model with FD preprocessing was significantly improved, with an accuracy of 96.68 % and a misclassification rate of only 3.32 %. The study effectively classifies transgenic corn with high accuracy, showing potential for widespread application in quality detection. © 2025 Elsevier Inc.

关键词： Classification Convolutional neural network Support vector machine Terahertz spectroscopy Transgenic corn

Corrigendum to “Classification of transgenic corn varieties using terahertz spectroscopy and convolutional neural network” [J. Food Compos. Anal. 145 (2025) 107771]

学校读者我要写书评

暂无评论

Journal of Food Composition and Analysis 2025年

作者： Yuying Jiang Xixi Wen Hongyi Ge Guangming Li Hao Chen Mengdie Jiang Qingcheng Sun Shilei Wei Peng Li Key Laboratory of Grain Information Processing and Control Henan University of Technology Ministry of Education Zhengzhou 450001 China Henan Provincial Key Laboratory of Grain Photoelectric Detection and Control Henan University of Technology Zhengzhou 450001 China School of Artificial Intelligence and Big Data Henan University of Technology Zhengzhou 450001 China School of Information Science and Engineering Henan University of Technology Zhengzhou 450001 China Institute for Complexity Science Henan University of Technology Zhengzhou 450001 China