检索结果-内蒙古大学图书馆

IEEE Workshop on Applications of Computer Vision (WACV)

作者： Thanh-Son Nguyen Hong Yang Basura Fernando Institute of High Performance Computing Agency for Science Technology and Research Singapore Centre for Frontier AI Research (CFAR) Agency for Science Technology and Research Singapore

ISBN: (数字)9798331510831

ISBN: (纸本)9798331510848

Annotating scene graphs for images is a time-consuming task, resulting in many instances of missing relations within existing datasets. In this paper, we introduce the Statistical Relation Distillation (SRD) method to enhance scene graph datasets. SRD leverages human-annotated relations alongside object-to-object and predicate-to-predicate similarities to reinforce the existence likelihood of scene graph relations. Moreover, SRD can augment relational frequency using relations of non-selected object and predicate categories that are usually omitted by scene graph generation (SGG) task. The output from SRD derives the prior probability which is combined with model-predicted probabilities to annotate missing relations in training images and subsequently re-train SGG models on the augmented dataset. We evaluate our proposed method on Visual Genome and GQA-200 datasets. Experimental results show that training on the augmented dataset enhances the performance of prominent scene-graph generation models. The implementation code is at https://***/LUNAProject22/SRD.

关键词： Training Measurement Visualization Computer vision Codes Genomics Tail Probability Transformers Bioinformatics

来源：评论

学校读者我要写书评

暂无评论

Deduce and Select Evidences with Language Models for Training-Free Video Goal Inference

Deduce and Select Evidences with Language Models for Trainin...

引用

2025 IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2025

作者： Ee, Yeo Keat Zhang, Hao Matyasko, Alexander Fernando, Basura Institute of High-Performance Computing Agency for Science Technology and Research 1 Fusionopolis Way #16-16 Connexis Singapore138632 Singapore Centre for Frontier Ai Research Agency for Science Technology and Research 1 Fusionopolis Way #16-16 Connexis Singapore138632 Singapore

ISBN: (纸本)9798331510831

We introduce ViDSE, a Video framework that Deduce and Selects visual Evidence for training-free video goal inference using language models. Unlike approaches that directly apply vision-language models (VLM) or combine VLM+LLM to process dense video visuals, ViDSE explicitly selects relevant visual evidence (e.g., frames) based on the hypothesis deduced by the LLM. This approach not only im-proves accuracy but also reveals the logical process behind the model's decisions, enhancing explainability. Our exper-iments demonstrate that this selection process significantly reduces ambiguity in the subsequent inference reasoning stage and outperforms VLM-only and VLM+LLM models on goal inference tasks such as CrossTask and COIN. We further validate ViDSE' s generalizability and robustness on action recognition tasks, such as ActivityNet and UCF-101, under training-free and open-vocabulary conditions. We observe that ViDSE easily generalizes to other video tasks (e.g., action recognition) requiring filtering of redundant and irrelevant information. © 2025 IEEE.

关键词： Unified Modeling Language

来源：评论

学校读者我要写书评

暂无评论

Deduce and Select Evidences with Language Models for Training-Free Video Goal Inference

Deduce and Select Evidences with Language Models for Trainin...

引用

IEEE Workshop on Applications of Computer Vision (WACV)

作者： Yeo Keat Ee Hao Zhang Alexander Matyasko Basura Fernando Institute of High-Performance Computing Agency for Science Technology and Research Singapore Centre for Frontier AI Research Agency for Science Technology and Research Singapore Republic of Singapore

ISBN: (数字)9798331510831

ISBN: (纸本)9798331510848

We introduce ViDSE, a Video framework that Deduce and Selects visual Evidence for training-free video goal inference using language models. Unlike approaches that directly apply vision-language models (VLM) or combine VLM+LLM to process dense video visuals, ViDSE explicitly selects relevant visual evidence (e.g., frames) based on the hypothesis deduced by the LLM. This approach not only im-proves accuracy but also reveals the logical process behind the model's decisions, enhancing explainability. Our exper-iments demonstrate that this selection process significantly reduces ambiguity in the subsequent inference reasoning stage and outperforms VLM-only and VLM+LLM models on goal inference tasks such as CrossTask and COIN. We further validate ViDSE’ s generalizability and robustness on action recognition tasks, such as ActivityNet and UCF-101, under training-free and open-vocabulary conditions. We observe that ViDSE easily generalizes to other video tasks (e.g., action recognition) requiring filtering of redundant and irrelevant information.

关键词： Visualization Computer vision Accuracy Filtering Large language models Computational modeling Robustness Cognition

来源：评论

学校读者我要写书评

暂无评论

Impact of Glyph Information on Latent Space Diffusion Models for Accurate Handwritten Text Generation

Impact of Glyph Information on Latent Space Diffusion Models...

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Ying-Li Lin Hao-Chung Cheng Chung-I Huang Chien-Yao Wang Jia-Ching Wang National Central University Taoyuan Taiwan National Center for High-Performance Computing Hsinchu Taiwan Institute of Information Science Academia Sinica Taipei Taiwan

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

The generation of high-quality stylized handwritten text images is a challenging task in computer vision and artificial intelligence. While advanced approaches using Latent Diffusion Models (LDMs) for generating stylized handwritten text have shown effectiveness, they often struggle with maintaining the structural integrity of certain characters, resulting in issues such as missing or extraneous strokes. In this work, we propose GlyphLDM, an innovative model that integrates glyph image information into both the diffusion and denoising processes in the latent space, enhancing the structural accuracy of generated text images. In the early training stages, our method demonstrated a significant improvement in the structural accuracy of the generated text images, with the Average Confidence Score increasing by approximately 40% compared to the baseline method. These experimental results indicate that incorporating glyph image information has promising potential to enhance the structural accuracy and overall quality of generated text images. This approach provides an effective solution for generating more accurate and diverse handwritten text images.

关键词： Training Measurement Handwriting recognition Accuracy Optical character recognition Noise reduction Diffusion processes Signal processing Diffusion models Speech processing

来源：评论

学校读者我要写书评

暂无评论

Reinforcement Learning for Automated Pragma-Based Loop Optimization

Reinforcement Learning for Automated Pragma-Based Loop Optim...

引用

IEEE Information technology and Mechatronics Engineering Conference (ITOEC)

作者： Jiaxuan Li Qilong Zheng School of Computer Science University of Science and Technology of China (USTC) Hefei China National High Performance Computing Center University of Science and Technology of China (USTC) Hefei China

ISBN: (数字)9798331529482

ISBN: (纸本)9798331529499

Loops often dominate the execution time in high-performance computing, effective loop optimization is critical for overall performance. We propose a reinforcement learning–based framework that automatically discovers and composes transformations—including tiling, fusion, interchange, and unrolling—and evaluate the framework on a subset of Polybench benchmarks. Compared to the Polly compiler baseline, our approach achieves an average speedup of 2.46×, peaking at 7× on the jacobi-1d kernel, while also consistently outperforming a global greedy scheduling algorithm. By adaptively combining multiple transformations, the RL-based method exploits deeper synergies with minimal overhead once trained, thus alleviating the repeated manual tuning and hardware-specific adjustments required by conventional techniques.

关键词： Productivity Mechatronics Scheduling algorithms high performance computing Reinforcement learning Manuals Benchmark testing Kernel Optimization Tuning

来源：评论

学校读者我要写书评

暂无评论

ViLa-MIL: Dual-scale Vision-Language Multiple Instance Learning for Whole Slide Image Classification

arXiv

引用

arXiv 2025年

作者： Shi, Jiangbo Li, Chen Gong, Tieliang Zheng, Yefeng Fu, Huazhu School of Computer Science and Technology Xi’an Jiaotong University Xi’an China Jarvis Research Center Tencent YouTu Lab Shenzhen China Institute of High Performance Computing Agency for Science Technology and Research Singapore

Multiple instance learning (MIL)-based framework has become the mainstream for processing the whole slide image (WSI) with giga-pixel size and hierarchical image context in digital pathology. However, these methods heavily depend on a substantial number of bag-level labels and solely learn from the original slides, which are easily affected by variations in data distribution. Recently, vision language model (VLM)-based methods introduced the language prior by pre-training on large-scale pathological image-text pairs. However, the previous text prompt lacks the consideration of pathological prior knowledge, therefore does not substantially boost the model’s performance. Moreover, the collection of such pairs and the pre-training process are very time-consuming and source-intensive. To solve the above problems, we propose a dual-scale vision-language multiple instance learning (ViLa-MIL) framework for whole slide image classification. Specifically, we propose a dual-scale visual descriptive text prompt based on the frozen large language model (LLM) to boost the performance of VLM effectively. To transfer the VLM to process WSI efficiently, for the image branch, we propose a prototype-guided patch decoder to aggregate the patch features progressively by grouping similar patches into the same prototype;for the text branch, we introduce a context-guided text decoder to enhance the text features by incorporating the multi-granular image contexts. Extensive studies on three multi-cancer and multi-center subtyping datasets demonstrate the superiority of ViLa-MIL. Copyright © 2025, The Authors. All rights reserved.

关键词： Visual languages

来源：评论

学校读者我要写书评

暂无评论

Understanding EFX Allocations: Counting and Variants

arXiv

引用

arXiv 2025年

作者： Neoh, Tzeh Yuan Teh, Nicholas Institute of High Performance Computing Agency for Science Technology and Research Singapore Centre for Frontier AI Research Agency for Science Technology and Research Singapore University of Oxford United Kingdom

Envy-freeness up to any good (EFX) is a popular and important fairness property in the fair allocation of indivisible goods, of which its existence in general is still an open question. In this work, we investigate the problem of determining the minimum number of EFX allocations for a given instance, arguing that this approach may yield valuable insights into the existence and computation of EFX allocations. We focus on restricted instances where the number of goods slightly exceeds the number of agents, and extend our analysis to weighted EFX (WEFX) and a novel variant of EFX for general monotone valuations, termed EFX+. In doing so, we identify the transition threshold for the existence of allocations satisfying these fairness notions. Notably, we resolve open problems regarding WEFX by proving polynomial-time computability under binary additive valuations, and establishing the first constant-factor approximation for two agents. Copyright © 2025, The Authors. All rights reserved.

关键词： Polynomial approximation

来源：评论

学校读者我要写书评

暂无评论

NHC-mediated photocatalytic para-selective C–H acylation of aryl alcohols: regioselectivity control via remote radical spiro cyclization

引用

Science China Chemistry 2025年

作者： Tinglei Zhang Lala Wang Xiaolin Peng Tianhui Liao Dongmei Chen Ting Tu Donghan Liu Ziye Cheng Shiqing Huang Shi-Chao Ren Xinglong Zhang Yonggui Robin Chi State Key Laboratory of Green Pesticide Key Laboratory of Green Pesticide and Agricultural Bioengineering Ministry of Education Guizhou University Qiannan Medical College for Nationalities School of Chemistry Chemical Engineering and Biotechnology Nanyang Technological University Department of Chemistry The Chinese University of Hong Kong Institute of High Performance Computing Agency for Science Technology and Research

Friedel-Crafts reaction and transition-metal catalysis represent primary methodologies for arene acylation; however, suffer from certain limitations, such as electronic effect-dependent reactivity, regioselectivity issues, the necessity of directing group, and relatively exceptionally limited exploration in meta-and para-acylations. N-heterocyclic carbene(NHC) catalysis has recently emerged as a promising alternative strategy for arene acylation, offering distinct regioselectivity profiles that are challenging to achieve through known methods. Herein, an NHC-mediated photo-catalytic para-selective C–H acylation of aryl alcohols is disclosed. The reaction employs a remote radical-induced temporary spiro cyclization strategy, proceeding via tandem radical dearomatization, radical acylation, rearomatization processes. Combined experimental and computational mechanistic studies support the proposed reaction pathway and observe an exclusive para-selectivity regardless of the nature of the substituents on the substrates. Despite the less satisfactory yields, the current method exhibits non-electronic nature-based reactivity and regioselectivity, enabling quick access to acylated arenes that are difficult or impossible for reported methods.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Robust projection parameter calibration in cryo-ET with L1-norm optimization

引用

Ultramicroscopy 2025年 274卷 114134页

作者： Guo, Shengkai Xu, Zihe Li, Xinyan Yang, Zhidong Feng, Chenjie Han, Renmin Cheeloo College of Medicine Shandong University Qingdao266237 China High Performance Computer Research Center Institute of Computing Technology CAS Beijing100190 China College of Medical Information and Engineering Ningxia Medical University Yinchuan750004 China

Fiducial marker-based alignment in cryo-electron tomography (cryo-ET) has been extensively studied over a long period. The calibration of projection parameters using nonlinear least squares technique methodologies stands as the ultimate and pivotal stage in the alignment procedure. The efficacy of calibration is substantially impacted by noise and outliers in the marker data obtained from previous steps. Several robust fitting methods have been explored and implemented to address this issue by improving marker data or assigning weights to markers. However, these methods have their own limitations and often assume general Gaussian noise assumption, which may not accurately represent the distribution of noise and outliers in the marker data. In this work, we propose a robust projection parameter calibration model based on L1-norm optimization under Laplace noise assumption in order to overcome the limitations of existing methods. To efficiently solve the problem, we also design a faster and stabler first-order non-sparse method based on smooth approximation strategy. Additionally, we introduce subgradient and subdifferential for mathematical analysis. The accuracy, robustness, and efficacy of our approach are demonstrated through both simulated and real-world experiments. © 2025 Elsevier B.V.

关键词： Least squares approximations

来源：评论

学校读者我要写书评

暂无评论

Testing eigenstate thermalization hypothesis on small scale quantum processors

arXiv

引用

arXiv 2025年

作者： Srihari, Maanav Shaji, Anil School of Physics Indian Institute of Science Education and Research Kerala Thiruvananthapuram695551 India Center for High Performance Computing Indian Institute of Science Education and Research Kerala Thiruvananthapuram695551 India

The Eigenstate Thermalization Hypothesis (ETH) is a framework for discussing thermal behavior originating from chaotic dynamics in isolated many-body quantum systems. The PXP model, where certain states do not thermalize, has been compared with the Sachdev-Ye Kitaev (SYK) model, which is believed to be fully thermalizing. A gate-based quantum circuit approach is utilized to simulate time evolution and compute the Out-of-Time-Ordered Correlator (OTOC), a measure of the extent of chaos. Considering restrictions on implementing SYK on gate-based hardware, a simplified model called Spin-XY4 (SXY4), which has a thermal behavior similar to SYK, is tested. An alternate method, which optimizes control on an analog quantum device with the GRAPE (GRadient Ascent Pulse Engineering) algorithm, is also utilized to simulate the SYK model. © 2025, CC BY.

关键词： Quantum electronics

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：