检索结果-内蒙古大学图书馆

36th IEEE International symposium on Defect and Fault Tolerance in VLSI and Nanotechnology Systems, DFT 2023

作者： Grignani, Wesley Santos, Douglas A. Dilillo, Luigi Viel, Felipe Melo, Douglas R. University of Montpellier Cnrs Ies Montpellier France University of Vale Do Itajaí Leds Itajaí Brazil Federal University of Santa Catarina SpaceLab Florianópolis Brazil

ISBN: (纸本)9798350315004

Several remote sensing applications that collect specific data in the space environment use images capable of providing a large volume of information, known as hyperspectral images. Given the amount of data, one of the most critical issues in applications that use hyperspectral images is the demanded for compression, which also affects restrictions on the storage capacity and processing in space applications. This work aimed to implement hyperspectral image compressors, considering the standard by the CCSDS (Consultative Committee for Space Data Systems). The solutions were implemented using a High-Level Synthesis tool (HLS) and a manual description in a Hardware Description Language (HDL). Results show that, compared to the software solution, the HLS and the HDL implementation accelerated the application by 1.6× and 4×, respectively. For images up to 512 spectral bands, the HLS solution presented a throughput of 9.11 MSa/s, while the HDL solution can process 21.47 MSa/s, which meets the real-time requirements of the standard. The HDL solution uses about 3× fewer LUTs (Look-Up Tables) LUTs and 8× fewer FFs (Flip-Flops) than the HLS implementation. Due to the low cost observed in the results, we intend to harden the accelerator and integrate it into a future multi-core satellite system. © 2023 IEEE.

关键词： Table lookup

来源：评论

学校读者我要写书评

暂无评论

Thermographic Non-Invasive Inspection Modelling of Fertilizer Pipelines Using Neural Networks 33

Thermographic Non-Invasive Inspection Modelling of Fertilize...

引用

33rd SIBGRAPI Conference on graphics, Patterns and images (SIBGRAPI)

作者： Duarte, Marta Coch, Victor Dias, Jovania Botelho, Silvia Duarte, Nelson Drews, Paulo Fed Univ Rio Grande FURG Comp Sci Ctr C3 Rio Grande Brazil

ISBN: (纸本)9781728192741

Industry pipeline fault, like blockage can create major problems for engineers and financial loss for the company. The blockage detection is necessary for smooth functioning of an industry and safety of the environment. This work presents a model for non-invasive inspection of pipes. It proposes the use of a neural network to identify the obstruction stage in fertilizer industry, using external thermal images obtained from the pipelines. A dataset capable of mapping the external thermal behavior in profile of the internal deposit is developed. The Multilayer Perceptron neural network was able to learn the thermal pixel mapping in a deposit profile, obtaining satisfactory results.

关键词： Fertilizer Neural Networks Pipeline Inspection Thermal image

来源：评论

学校读者我要写书评

暂无评论

Contrastive Learning and Iterative Meta-Pseudo-Labeling on 2D Projections for Deep Semi-Supervised Learning

Contrastive Learning and Iterative Meta-Pseudo-Labeling on 2...

引用

brazilian symposium on computer graphics and image processing (SIBGRAPI)

作者： David Aparco-Cardenas Jancarlo F. Gomes Alexandre X. Falcão Pedro J. De Rezende Institute of Computing (IC) University of Campinas (UNICAMP) Campinas Brazil

ISBN: (数字)9798350376036

ISBN: (纸本)9798350376043

The scarcity of accurately labeled data critically hampers the usage of deep learning models. This issue is highlighted in areas (e.g., biological sciences) where data annotation results in an expert-demanding, labor-intensive and error-prone task. While state-of-the-art semi-supervised approaches have proven effective in circumventing this limitation, their reliance on pre-trained architectures and large validation sets to deliver effective solutions still poses a challenge. In this work we introduce an iterative contrastive-based meta-pseudo-Iabeling method for training non-pre-trained custom CNN architectures for image classification in conditions of limited labeled and abundant unlabeled data, with no dependency on a validation set. It generates multiple models across a few iterations, which are in turn exploited in an ensemble manner to label the unlabeled data and train a final classifier. Our approach starts by capitalizing on contrastive learning to enhance the representation ability of two collaborative networks while eliminating the need of pre-trained architectures. Then, during each iteration, the networks are trained within a teacher-student based cross-training setup, where OPFSemi (teacher) propagates labels from labeled to unlabeled on the non-linear 2D latent space projections of each network's (student) deep features; afterward, the pseudo-labels with the highest top 10% confidence, per class, are picked to fine-tune the other network in a cross-training manner, jointly mitigating confirmation bias and overfitting while improving the generalization ability of the networks as iterations evolve. Our method is evaluated on three challenging biological image datasets with only 5 % of labeled samples, demonstrating its effectiveness and robustness when compared to two direct baselines and six state-of-the-art methods from three different semi-supervised learning paradigms.

关键词： Training Federated learning Annotations Contrastive learning Semisupervised learning Robustness Data models Iterative methods Labeling image classification

来源：评论

学校读者我要写书评

暂无评论

Faster and Accurate Compressed Video Action Recognition Straight from the Frequency Domain 33

Faster and Accurate Compressed Video Action Recognition Stra...

引用

33rd SIBGRAPI Conference on graphics, Patterns and images (SIBGRAPI)

作者： dos Santos, Samuel Felipe Almeida, Jurandy Univ Fed Sao Paulo UNIFESP Inst Ciencia & Tecnol BR-12247014 Sao Jose Dos Campos SP Brazil

ISBN: (纸本)9781728192741

Human action recognition has become one of the most active field of research in computer vision due to its wide range of applications, like surveillance, medical, industrial environments, smart homes, among others. Recently, deep learning has been successfully used to learn powerful and interpretable features for recognizing human actions in videos. Most of the existing deep learning approaches have been designed for processing video information as RGB image sequences. For this reason, a preliminary decoding process is required, since video data are often stored in a compressed format. However, a high computational load and memory usage is demanded for decoding a video. To overcome this problem, we propose a deep neural network capable of learning straight from compressed video. Our approach was evaluated on two public benchmarks, the UCF-101 and HMDB-51 datasets, demonstrating comparable recognition performance to the state-of-the-art methods, with the advantage of running up to 2 times faster in terms of inference speed.

关键词： action recognition compressed domain processing convolutional neural network frequency domain

来源：评论

学校读者我要写书评

暂无评论

From superpixels to foundational models: An overview of unsupervised and generalizable image segmentation

引用

computerS & graphics-UK 2024年 123卷

作者： Rodrigues, Cristiano N. Nunes, Ian M. Pereira, Matheus B. Oliveira, Hugo dos Santos, Jefersson A. Univ Fed Minas Gerais UFMG Dept Comp Sci DCC BR-31270901 Belo Horizonte MG Brazil Brazilian Inst Geog & Stat IBGE Stat Dept DPE BR-20031170 Rio De Janeiro Brazil Univ Fed Vicosa UFV Dept Comp Sci DPI BR-36570900 Vicosa Brazil Univ Sheffield Dept Comp Sci Sheffield S1 4DP England

image segmentation is one of the most classical computer vision tasks. Segmentation tasks yield a set of classes attributed to individual pixels instead of sparsely predicted images or patches, such as in classification or detection tasks. However, creating annotation sets for pixelwise tasks is a very costly task, often requiring hours for labeling single samples in images with multiple classes of objects. In this context, unsupervised learning can be leveraged either to expedite the annotation procedure and/or to guide the segmentation algorithms altogether without the need for manual annotations. Classical unsupervised segmentation methods leveraged techniques from areas as graph theory, image processing, clustering or supervised classifiers in order to achieve "shallow"pixelwise classification. These techniques usually aim to achieve superpixel oversegmentations by grouping similar pixels that should pertain to the same object. Modern deep unsupervised approaches for image segmentation aimed to group pixels in a data-driven way by using the capabilities of deep architectures to process unstructured data such as images. Later, self-supervised learning bypassed the need for labels via pretext tasks, compelling deep architectures to learn more generic features capable of enhancing downstream tasks, including segmentation. The generalized representations produced by unsupervised models have propelled the recent progress in self-supervised, few- and zero-shot learning and even general-purpose foundational models in computer vision, yielding state-of-the-art results across diverse tasks and datasets. This paper provides an overview of unsupervised and generalizable approaches for image segmentation, introduces key concepts and terminology, and discusses the main aspects of state-of-the-art methods. Additionally, we highlight prominent applications in various domains such as remote sensing, medical imaging, and geology. Finally, we discuss trends and future directions

关键词： image segmentation Unsupervised learning Deep learning Superpixel Representation learning Zero-shot

来源：评论

学校读者我要写书评

暂无评论

A Bucket LBVH Construction and Traversal Algorithm for Volumetric Sparse Data 33

A Bucket LBVH Construction and Traversal Algorithm for Volum...

引用

33rd SIBGRAPI Conference on graphics, Patterns and images (SIBGRAPI)

作者： Fernandes, I. B. Walter, M. Univ Fed Rio Grande do Sul Inst Informat PPGC Porto Alegre RS Brazil

ISBN: (纸本)9781728192741

Many volumetric rendering algorithms use spatial 3D grids as the underlying data structure. Efficient representation, construction, and traversal of these grids are essential to achieve real-time performance, particularly for time-varying data such as in fluid simulations. In this paper, we present improvements on algorithms for building and traversing Bounding Volume Hierarchies (BVH) designed for sparse volumes. Our main insight was to simplify data layout representation by grouping voxels in buckets, preserving their spatiality using Morton codes, instead of using bricks, as current solutions use. Our solution does not use pointers nor stacks, allowing for its usage directly on computing shaders and provides, on average, 9.3x improvement in construction speed, compared with stateof-the-art approaches for Linear Bounding Volume Hierarchies (LBVH).

关键词： Three dimensional computer graphics

来源：评论

学校读者我要写书评

暂无评论

Investigating the Cognitive Differences in Scene processing and Aesthetic Evaluation When Viewing Human- vs. AI-generated imagery 25

Investigating the Cognitive Differences in Scene Processing ...

引用

Proceedings of the 2025 symposium on Eye Tracking Research and Applications

作者： Long Feng Huang Psychology University of Toronto Toronto Ontario Canada

ISBN: (纸本)9798400714870

The aesthetic experience of scene perception can be influenced by a variety of cognitive, perceptual, and oculomotor factors. Using a passive viewing eye-tracking paradigm, we explored the differences in visuospatial attention for free vs. guided viewing of human- vs. AI-generated scene photographs and paintings. Aesthetic ratings and fixation data were collected for each image, and heatmaps were generated to determine the spatial distribution of gaze. This work has the potential to reveal crucial differences in the visual processing and subsequent aesthetic evaluation of scene images of various sources and styles, and characterize what factors – if any – differentiate AI images from human creations.

关键词： AI art

来源：评论

学校读者我要写书评

暂无评论

Computationally-Efficient Neural image Compression Using Asymmetric Autoencoders

Computationally-Efficient Neural Image Compression Using Asy...

引用

symposium on Integrated Circuits and Systems Design (SBCCI)

作者： Leonardo Augusto Thiago L. T. Da Silveira Mateus Grellert Federal University of Rio Grande do Sul (UFRGS) Porto Alegre Brazil

ISBN: (数字)9798350391695

ISBN: (纸本)9798350391701

Neural image compression (NIC) is an emerging field, with notable works achieving promising results in image quality. However, these networks introduce new challenges regarding computational cost since NIC require high computational power for compression and decompression, demanding high processing time even when specialized platforms like GPUs are used. This work proposes optimizing one of the reference models available in the literature to achieve better processing time while maintaining the compression quality. Unlike other solutions that aim at higher compression or better image quality, our proposal seeks to reduce the computational cost of NIC techniques. This is achieved by designing novel asymmetric autoencoder architectures with cost-efficient versions of the decoder component of these networks. Experimental results obtained with the Kodak dataset show that, by pruning layers or kernels at the decoder network, it is possible to achieve a 18% reduction in decompression time for GPU, with an average PSNR loss of 0.001 dB. When tested on a CPU, the decompression time reduction was 46%, with a PSNR loss of 0.007 dB. We expect that our contributions will pave the way for more efficient NIC techniques, making them more accessible for real-time applications and resource-constrained environments.

关键词： image quality image coding graphics processing units computer architecture Throughput Decoding Computational efficiency Central processing Unit Proposals System analysis and design

来源：评论

学校读者我要写书评

暂无评论

Performance and error analysis of recursive edge-aware Gaussian filters on GPUs 33

Performance and error analysis of recursive edge-aware Gauss...

引用

33rd SIBGRAPI Conference on graphics, Patterns and images (SIBGRAPI)

作者： Ferreira, Hermes H. Gastal, Eduardo S. L. Schnorr, Lucas M. Navaux, Philippe O. A. Univ Fed Rio Grande do Sul Inst Math & Stat Porto Alegre RS Brazil Univ Fed Rio Grande do Sul Inst Informat Porto Alegre RS Brazil

ISBN: (纸本)9781728192741

We present a schematic for image edge-aware Gaussian GPU filtering which has linear complexity on the number of pixels of the image. It allows us to reduce the execution time as we increase the number of Streaming Multiprocessors (SMs) on the GPU. We make use of a domain transformation and use a complex-valued recursive formulation of the Gaussian filter. The algorithm partitions the image in disjoint regions, where we compute in parallel the filtering operations, avoiding communication between regions. Our implementation leads to a real-time solution using a modern GPU. With the RTX 2080 Ti, we achieved an execution time of less than 10 milliseconds for 2 filtering iterations on high-resolution RGB images of dimensions 2048 x 2048.

关键词： edge aware filtering GPU processing high performance computing image processing recursive filtering

来源：评论

学校读者我要写书评

暂无评论

Non-Rigid Registration of Time-Lapse Cell Microscopy images Improved by a Semi-Incremental Optimization Method

Non-Rigid Registration of Time-Lapse Cell Microscopy Images ...

引用

IEEE International symposium on Biomedical Imaging

作者： Qi Gao Karl Rohr Core Facility Platform Mannheim Medical Faculty Mannheim Heidelberg University Germany Biomedical Computer Vision Group BioQuant IPMB Heidelberg University Germany

ISBN: (数字)9798331520526

ISBN: (纸本)9798331520533

Non-rigid registration of cell nuclei in time-lapse microscopy images needs accurate estimation of the deformation fields between the reference and all other images. To address the issue of accumulated errors in classical temporal incremental registration approaches, we introduce a new semi-incremental optimization method. Based on a proper initialization constructed through motion concatenation, which exploits temporal coherence within the image sequence, the deformation field between each image and the reference is computed, reducing accumulated errors and yielding more reliable results. Experiments on real time-lapse cell images demonstrate that our method outperforms previous approaches, including deep learning models, in terms of registration accuracy. Additionally, computation on a GPU significantly enhances efficiency.

关键词： image registration Accuracy Deformation Microscopy Optimization methods graphics processing units Estimation Fluorescence image sequences Reliability

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：