检索结果-内蒙古大学图书馆

12th International Conference on Learning Representations, ICLR 2024

作者： Cao, Chenjie Ren, Xinlin Fu, Yanwei School of Data Science Fudan University China Shanghai Key Lab of Intelligent Information Processing School of Computer Science Fudan University China

Recent advancements in learning-based Multi-View Stereo (MVS) methods have prominently featured transformer-based models with attention mechanisms. However, existing approaches have not thoroughly investigated the profound influence of transformers on different MVS modules, resulting in limited depth estimation capabilities. In this paper, we introduce MVSFormer++, a method that prudently maximizes the inherent characteristics of attention to enhance various components of the MVS pipeline. Formally, our approach involves infusing cross-view information into the pre-trained DINOv2 model to facilitate MVS learning. Furthermore, we employ different attention mechanisms for the feature encoder and cost volume regularization, focusing on feature and spatial aggregations respectively. Additionally, we uncover that some design details would substantially impact the performance of transformer modules in MVS, including normalized 3D positional encoding, adaptive attention scaling, and the position of layer normalization. Comprehensive experiments on DTU, Tanks-and-Temples, BlendedMVS, and ETH3D validate the effectiveness of the proposed method. Notably, MVSFormer++ achieves state-of-the-art performance on the challenging DTU and Tanks-and-Temples benchmarks. Codes and models are available at https://***/maybeLx/MVSFormerPlusPlus. © 2024 12th International Conference on Learning Representations, ICLR 2024. All rights reserved.

关键词： Benchmarking

来源：评论

学校读者我要写书评

暂无评论

Standing on the Shoulders of Giants: Reprogramming Visual-Language Model for General Deepfake Detection 39

Standing on the Shoulders of Giants: Reprogramming Visual-La...

引用

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Lin, Kaiqing Lin, Yuzhen Li, Weixiang Yao, Taiping Li, Bin Guangdong Provincial Key Laboratory of Intelligent Information Processing Shenzhen Key Laboratory of Media Security SZU-AFS Joint Innovation Center for AI Technology Shenzhen University Shenzhen518060 China Tencent Youtu Lab China

ISBN: (纸本)157735897X

The proliferation of deepfake faces poses huge potential negative impacts on our daily lives. Despite substantial advancements in deepfake detection over these years, the generalizability of existing methods against forgeries from unseen datasets or created by emerging generative models remains constrained. In this paper, inspired by the zero-shot advantages of Vision-Language Models (VLMs), we propose a novel approach that repurposes a well-trained VLM for general deepfake detection. Motivated by the model reprogramming paradigm that manipulates the model prediction via input perturbations, our method can reprogram a pre-trained VLM model (e.g., CLIP) solely based on manipulating its input without tuning the inner parameters. First, learnable visual perturbations are used to refine feature extraction for deepfake detection. Then, we exploit information of face embedding to create sample-level adaptative text prompts, improving the performance. Extensive experiments on several popular benchmark datasets demonstrate that (1) the cross-dataset and cross-manipulation performances of deepfake detection can be significantly and consistently improved (e.g., over 88% AUC in cross-dataset setting from FF++ to WildDeepfake);(2) the superior performances are achieved with fewer trainable parameters, making it a promising approach for real-world applications. © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Disentangle Nighttime Lens Flares: Self-supervised Generation-based Lens Flare Removal 39

Disentangle Nighttime Lens Flares: Self-supervised Generatio...

引用

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： He, Yuwen Wang, Wei Wu, Wanyu Jiang, Kui Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System Wuhan University of Science and Technology China School of Computer Science and Technology Harbin Institute of Technology China

ISBN: (纸本)157735897X

Lens flares arise from light reflection and refraction within sensor arrays, whose diverse types include glow, veiling glare, reflective flare and so on. Existing methods are specialized for one specific type only, and overlook the simultaneous occurrence of multiple typed lens flares, which is common in the real-world, e.g. coexistence of glow and displacement reflections from the same light source. These co-occurring lens flares cannot be effectively resolved by the simple combination of individual flare removal methods, since these coexisting flares originates from the same light source and are generated simultaneously within the same sensor array, exhibit a complex interdependence rather than simple additive relation. To model this interdependent flares’ relationship, our Nighttime Lens Flare Formation model is the first attempt to learn the intrinsic physical relationship between flares on the imaging plane. Building on this physical model, we introduce a solution to this joint flare removal task named Self-supervised Generation-based Lens Flare Removal Network (SGLFR-Net), which is self-supervised without pre-training. Specifically, the nighttime glow is detangled in PSF Rendering Network(PSFR-Net) based on PSF Rendering Prior, while the reflective flare is modelled in Texture Prior Based Reflection Flare Removal Network (TPRR-Net). Empirical evaluations demonstrate the effectiveness of the proposed method in both joint and individual glare removal tasks. © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词： Luminescent devices

来源：评论

学校读者我要写书评

暂无评论

Learned Fingerprint Embedding for Large-Scale Peptide Mass Spectra Retrieval

Learned Fingerprint Embedding for Large-Scale Peptide Mass S...

引用

2023 IEEE International Conference on Bioinformatics and Biomedicine, BIBM 2023

作者： Wang, Yongshuai Cai, Xiaojun Li, Defeng Sun, Shiwei Chen, Cheng Cui, Xuefeng Shandong University School of Computer Science and Technology Qingdao China Chinese Academy of Sciences Key Lab of Intelligent Information Processing Institute of Computing Technology Beijing China

ISBN: (纸本)9798350337488

Tandem mass spectrometry (MS/MS) is a widely used technique for protein identification, post-translational modifications, immunotherapy, and other applications. As the amount of MS/MS spectra data increases, new computational methods are needed to efficiently search through these databases. This study introduces MS2VEC, a novel fingerprint embedding model designed to facilitate large-scale retrieval of peptide mass spectra. MS2VEC captures the relationships between distant peaks and incorporates position-aware fingerprint features from all peaks. To do this, dilated convolutions are used to capture remote relationships, and a novel position-aware multi-head attention pooling mechanism is used to abstract fingerprint features. The results demonstrate that MS2VEC achieves a top-1 retrieval accuracy of 0.810, outperforming existing methods by 5.1%. Interestingly, the precursor charge is not essential for the retrieval task, as the spectra itself contains enough information to accurately predict the charge. Additionally, the results suggest that weight-balanced fragment ions and water losses are important contributors to fingerprint features. © 2023 IEEE.

关键词： deep learning embedding information retrieval peptide mass spectra tandem mass spectrometry

来源：评论

学校读者我要写书评

暂无评论

Underwater Image Restoration Based on Improved SwinIR Model 13

Underwater Image Restoration Based on Improved SwinIR Model

引用

13th International Conference on intelligent Control and information processing, ICICIP 2025

作者： Lu, Yanke Li, Fang Luo, Xiaonan Guangxi Key Laboratory of Image and Graphic Intelligent Processing Guilin University of Electronic Technology Guilin541004 China School of Computer Science and Information Security Guilin University of Electronic Technology Guilin541004 China Guilin541004 China Guangxi Key Laboratory of Image and Graphic Intelligent Processing Guilin541004 China

ISBN: (纸本)9798331516147

Image restoration is a classic foundational visual task, aimed at recovering damaged images, such as those affected by compression, blurring, or noise, to high-definition clarity. Although current image enhancement techniques largely rely on convolutional neural networks, the efficient capabilities of Transformers demonstrated in complex visual tasks have not yet been extensively applied in the field of image restoration. This paper presents an optimized image restoration strategy based on the SwinIR model, which is applied to the task of underwater image restoration. In the shallow feature extraction phase, deep separable convolution technology is employed;while in the deep feature extraction phase, a Spatial Gated Feedforward Network (SGFN) is utilized. The experimental data suggests that the novel approach surpasses conventional techniques in effectiveness and has effectively minimized both the model's parameter quantity and computational demands. © 2025 IEEE.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

Distributed Multiagent Resource Allocation Based on Transformer and DRL for Cloud XR Hybrid Content Transmission

引用

IEEE Internet of Things Journal 2025年第12期12卷 22110-22127页

作者： Wang, Zhaocheng Wu, Jun Wang, Rui Li, Ying Tongji University School of Computer Science and Technology Shanghai201804 China Fudan University School of Computer Science Shanghai Key Laboratory of Intelligent Information Processing Shanghai200433 China Tongji University College of Electronics and Information Engineering Shanghai Institute of Intelligent Science and Technology Shanghai201804 China Peng Cheng Laboratory Department of Strategic and Advanced Interdisciplinary Research Guangdong518055 China

Extended reality (XR) technologies and applications have grown rapidly in recent years. In addition to providing immersive ultrahigh-definition (UHD) video, XR also allows for a haptic experience where devices can be remotely manipulated to accomplish tasks. However, varying numbers of XR users accessing the communication system can strain limited spectrum resources, posing challenges in resource allocation. Therefore, this article studies resource blocks (RBs) allocation problem in a downlink transmission scenario where real-time cloud XR video and haptic contents need to be transmitted simultaneously. We also consider the random variation in the number of XR users and propose an adaptive distributed multiagent deep reinforcement learning (DRL) combined with Transformer (ADMA-DcT) for dynamic RB allocation method. This method addresses the dynamic change in network input dimensions due to user number variability using a state division module and a self-attention mechanism in the encoder module. To our knowledge, this is the first work to study the RB allocation problem in Cloud XR transmission considering simultaneous transmitting of video and haptic services with a dynamically changing user base. Our extensive simulations show that the ADMA-DcT model, end-to-end trained, outperforms other benchmarks in successfully serving a larger number of XR users under varying user number conditions, demonstrating excellent adaptivity and robustness. © 2014 IEEE.

关键词： Multi agent systems

来源：评论

学校读者我要写书评

暂无评论

CFinDEE: A Chinese Fine-Grained Financial Dataset for Document-Level Event Extraction 24

CFinDEE: A Chinese Fine-Grained Financial Dataset for Docume...

引用

33rd Companion of the ACM World Wide Web Conference, WWW 2023

作者： Zhang, Tian Liu, Maofu Zhou, Bingying School of Computer Science and Technology Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System Wuhan University of Science and Technology Wuhan China

ISBN: (纸本)9798400701726

Document-level event extraction faces numerous challenges in accurately modeling real-world financial scenarios, particularly due to the inadequacies in existing datasets regarding data scale and fine-grained annotations. The development of datasets is a crucial factor in driving research progress;therefore, we present a high-quality Chinese document-level event extraction dataset, CFinDEE. This dataset, grounded in real-world financial news, defines 22 event types and 116 argument roles, annotating 26,483 events and 107,096 event arguments. CFinDEE aims to address these shortcomings by providing more comprehensive annotations and data augmentation, offering richer resources for document-level event extraction in the financial domain. CFinDEE extends data both horizontally and vertically, where horizontal expansion enriches the types of financial events, enhancing the diversity of the dataset;vertical expansion, by increasing the scale of the data, effectively boosts the practical value of the dataset. Experiments conducted on multiple advanced models have validated the high applicability and effectiveness of the CFinDEE dataset for document-level event extraction tasks in the financial field. © 2024 Copyright held by the owner/author(s). Publication rights licensed to ACM.

关键词： Expansion

来源：评论

学校读者我要写书评

暂无评论

The Methodology of Mask Detection Employing the YOLOv7 Framework 4

The Methodology of Mask Detection Employing the YOLOv7 Frame...

引用

4th International Conference on Neural Networks, information and Communication Engineering, NNICE 2024

作者： Zhang, Fucheng Tian, Pinfang Wuhan University of Science and Technology School of Computer Science and Technology Wuhan China Hubei Key Laboratory of Intelligent Information Processing and Real-TimeIndustrial Systems Wuhan China

ISBN: (纸本)9798350394375

In the current field of object detection, the YOLOv7 model demonstrates significant advantages in terms of both detection speed and accuracy. However, the model shows deficiencies in focussing on key information, and its loss function's sensitivity to bounding boxes results in suboptimal performance in densely populated environments and small target detection scenarios. To address these issues, this study proposes a mask target detection method based on YOLOv7. By integrating a channel attention mechanism into the YOLOv7 model, this method enhances the model's focus on important information in images, effectively reducing interference from background factors. Additionally, the method incorporates Normalized Wasserstein Distance (NWD) into the loss function to optimize the model's evaluation of small target detection. Experimental verification shows that the improved model maintains its original detection speed while increasing the overall accuracy to 92.1%, a 2.9% improvement over the original YOLOv7 model. It achieves a 3.3% increase in the mAP.5 metric. This indicates that the optimised YOLOv7 model has superior detection capabilities, allowing it to perform mask-wearing detection tasks more rapidly and accurately in natural environments. © 2024 IEEE.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Personalized Federated Learning with Neighbor Aggregation for Traffic Flow Prediction 9

Personalized Federated Learning with Neighbor Aggregation fo...

引用

9th International Conference on Computer and Communications, ICCC 2023

作者： Guo, Junxin Feng, Xinxin Zheng, Haifeng Fuzhou University Fujian Key Lab for Intelligent Processing and Wireless Transmission of Media Information College of Physics and Information Engineering Fuzhou China

ISBN: (纸本)9798350317251

As an important part of intelligent Transportation System (ITS), traffic flow prediction has attracted more and more attention from academia and industry. However, since traffic flow data are distributed among different organizations, traditional centralized learning methods are no longer applicable, thus federated learning has begun to play an important role in traffic flow prediction tasks. In this paper, we design an improved traffic flow prediction model based on federated learning. At the same time, we consider the interaction of topology and feature space between traffic data owned by different organizations, and propose a method named Federated Neighbor Aggregation (FedNe). The server update the model parameters with both the physical connectivity and similarity relationships between the organizations, and finally generate a personalized model for each organization. Experimental results on real-world dataset show that our proposed method has the lowest average MAE and average RMSE. Finally, based on our method, when new clients participate, they can also quickly obtain their personalized models. © 2023 IEEE.

关键词： Forecasting

来源：评论

学校读者我要写书评

暂无评论

Towards Multi-label Open Set Diagnosis for Medical Images 22

Towards Multi-Label Open Set Diagnosis for Medical Images

引用

22nd IEEE International Symposium on Biomedical Imaging, ISBI 2025

作者： Pang, Shijie Zhao, Yuze Zhao, Rui-Wei Feng, Rui Academy for Engineering and Technology Fudan University Shanghai China School of Computer Science Fudan University Shanghai Key Laboratory of Intelligent Information Processing Shanghai China

ISBN: (纸本)9798331520526

In real clinical scenarios, diagnosis algorithms may face the complex concurrent presence of multiple common, rare, or even unknown disease conditions. In particular, rare disease categories may be underrepresented or even absent from the class-imbalanced training set, leaving safe and trustworthy diagnostic decision-making extremely challenging. In this work, we first formalize the medical image diagnosis as a multi-label open set recognition under class-imbalanced training data and propose a novel Multi-label Open set Diagnosis (MOSD) solution to address this problem. Our MOSD first employs a Balanced Multi-label Data Augmentation (BMDA) module to augment multi-label image data to alleviate the class imbalance in model training more effectively. Then it also boasts a Multi-label Evidential Deep Learning (MEDL) module with tailored learning objectives to better estimate the uncertainty for multi-label open set prediction. Experimental results on the ODIR-5K benchmark successfully demonstrate its superiority in both close set known class classification and open set unknown class detection. © 2025 IEEE.

关键词： Mammography

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：