检索结果-内蒙古大学图书馆

IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

作者： Yizhou Jin Yu Lu Gang Zhou Qingjie Liu Yunhong Wang State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing China Hangzhou Innovation Institute Beihang University

Glass wool defect detection is a key part of product quality assessment in the glass wool production process, yet few studies have been reported in this area. We propose a glass wool defect dataset named GWD, and also use the YOLOv5s model embedded in the GSConv and the CBAM modules for both Gap and Glueless defects in this dataset. The experimental results show that the performance of the improved YOLOv5s on the GWD dataset is superior to other compared methods and achieves a relatively good level on other publicly available datasets. Compared to the vanilla YOLOv5s, the mAP50 increased by 3.7% to 84.1%, the recall increased by 4.2% to 84.4%, and the number of parameters decreased by 0.42 MB to 6.27 MB of the improved YOLOv5s model on the GWD dataset. Speed-wisely, the improved YOLOv5s achieves a 97 FPS on RTX 2080Ti, thus making it practical to be applied in the industry of glass wool defect detection. The research on the GWD dataset is likely to contribute to breakthroughs in research on other datasets of the same type as well. The GWD dataset can be obtained by contacting us via email.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A Solution to Co-occurrence Bias: Attributes Disentanglement via Mutual Information Minimization for Pedestrian Attribute Recognition

arXiv

引用

arXiv 2023年

作者： Zhou, Yibo Hu, Hai-Miao Yu, Jinzuo Xu, Zhenbo Lu, Weiqing Cao, Yuran State key laboratory of virtual reality technology and systems Beihang University China Hangzhou Innovation Institute Beihang University China

Recent studies on pedestrian attribute recognition progress with either explicit or implicit modeling of the co-occurrence among attributes. Considering that this known a prior is highly variable and unforeseeable regarding the specific scenarios, we show that current methods can actually suffer in generalizing such fitted attributes interdependencies onto scenes or identities off the dataset distribution, resulting in the underlined bias of attributes co-occurrence. To render models robust in realistic scenes, we propose the attributes-disentangled feature learning to ensure the recognition of an attribute not inferring on the existence of others, and which is sequentially formulated as a problem of mutual information minimization. Rooting from it, practical strategies are devised to efficiently decouple attributes, which substantially improve the baseline and establish state-of-the-art performance on realistic datasets like PETAzs and RAPzs. © 2023, CC BY.

关键词：

来源：评论

学校读者我要写书评

暂无评论

HIPTrack: Visual Tracking with Historical Prompts

HIPTrack: Visual Tracking with Historical Prompts

引用

Conference on Computer Vision and Pattern Recognition (CVPR)

作者： Wenrui Cai Qingjie Liu Yunhong Wang State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing China Zhongguancun Laboratory Beijing China Hangzhou Innovation Institute Beihang University Hangzhou China

ISBN: (数字)9798350353006

ISBN: (纸本)9798350353013

Trackers that follow Siamese paradigm utilize similarity matching between template and search region features for tracking. Many methods have been explored to enhance tracking performance by incorporating tracking history to better handle scenarios involving target appearance variations such as deformation and occlusion. However, the utilization of historical information in existing methods is insufficient and incomprehensive, which typically requires repetitive training and introduces a large amount of computation. In this paper, we show that by providing a tracker that follows Siamese paradigm with precise and updated historical information, a significant performance improvement can be achieved with completely unchanged parameters. Based on this, we propose a historical prompt network that uses refined historical foreground masks and historical visual features of the target to provide comprehensive and precise prompts for the tracker. We build a novel tracker called HIPTrack based on the historical prompt network, which achieves considerable performance improvements without the need to retrain the entire model. We conduct experiments on seven datasets and experimental results demonstrate that our method surpasses the current state-of-the-art trackers on LaSOT, LaSOT ext , GOT-10k and NfS. Furthermore, the historical prompt network can seamlessly integrate as a plug-and-play module into existing trackers, providing performance enhancements. The source code is available at https://***/WenRuiCai/HIPTrack.

关键词： Training Visualization Computer vision Target tracking Accuracy Deformation Source coding

来源：评论

学校读者我要写书评

暂无评论

Algorithm for Acquiring Lettuce Plant Height Based on Image Recognition Network 4

Algorithm for Acquiring Lettuce Plant Height Based on Image ...

引用

4th International Conference on Artificial Intelligence and Electromechanical Automation, AIEA 2023

作者： Guo, Xiyue Zhong, Yong Zhao, Ming Zhang, Man Wang, Minjuan Key Laboratory of Modern Precision Agriculture System Integration Research Ministry of Education China Agricultural University China Yantai Institute of China Agricultural University Yantai264670 China State Key Laboratory of Virtual Reality Technology and Systems Beihang University China College of Information and Electrical Engineering China Agricultural University China

ISBN: (纸本)9781510666412

Lettuce height is one of the important phenotypic parameters of lettuce. Currently, evaluating plant height information based on RGB images is the most convenient and fastest method. However, these tasks often require complex camera calibration operations or reference objects with known heights, which are somewhat limiting. Therefore, we utilize a deep neural network to solve this problem to directly obtain the height of lettuce from RGB images without camera calibration and reference objects. This paper integrates four high-performance image recognition networks MobileNetV1, Densenet-121, ResNext-50, and EfficientNet-B3 using average fusion and linear fusion methods to predict lettuce plant height. Experiments show that we have achieved good results, the accuracy of the four basic models has been improved, and the average absolute error has been reduced to varying degrees. The smallest reduction is Efficient-Net, which has decreased 0.01 mm, the largest drop is the ResNext-50, which is 0.15mm lower. After the model is averagely fused, the average absolute error is further reduced to 1.22 mm, and the maximum absolute error is reduced to 6.48 mm, which is not only better than linear. The fusion results are better than the prediction results of a single model. © 2023 SPIE.

关键词： Lettuce

来源：评论

学校读者我要写书评

暂无评论

Efficient Sampling and Volume Rendering Strategy for Neural Field SLAM

Efficient Sampling and Volume Rendering Strategy for Neural ...

引用

IEEE International Conference on Multimedia and Expo (ICME)

作者： Meng Wang Yue Qi State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing China PengCheng Laboratory Shenzhen China Qingdao Research Institute of Beihang University Qingdao China

ISBN: (数字)9798350390155

ISBN: (纸本)9798350390162

Neural implicit representations have recently shown encouraging results in various domains, including promising progress in simultaneous localization and mapping (SLAM). Nevertheless, existing methods require substantial computational power (typically running on GPUs like RTX 3090) to maintain fast processing speeds and experience a noticeable slowdown on consumer-grade GPU devices. In this work, we present an efficient sampling and volume rendering strategy central to achieve faster processing speeds without additional memory costs. Two key ingredients enable our goal: (1) a progressive sampling method for tracking to perform efficient and robust camera poses estimated and (2) a feature fusion approach for color volume rendering to reduce calculation cost during the optimization process. Extensive experiments on four datasets (ScanNet, TUM RGB-D, Replica, and Synthetic) demonstrate that our approach can accelerate the optimization process and achieve better or no-par performance compared to the original method on a consumer-grade GPU device.

关键词： Performance evaluation Simultaneous localization and mapping Costs Three-dimensional displays Image color analysis Memory management Graphics processing units

来源：评论

学校读者我要写书评

暂无评论

Lightweight Spatial Embedding for Vision-based 3D Occupancy Prediction

arXiv

引用

arXiv 2024年

作者： Zhang, Jinqing Zhang, Yanan Liu, Qingjie Wang, Yunhong State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing China Zhongguancun Laboratory Beijing China Hangzhou Innovation Institute Beihang University Hangzhou China

Occupancy prediction has garnered increasing attention in recent years for its comprehensive fine-grained environmental representation and strong generalization to open-set objects. However, cumbersome voxel features and 3D convolution operations inevitably introduce large overheads in both memory and computation, obstructing the deployment of occupancy prediction approaches in real-time autonomous driving systems. Although some methods attempt to efficiently predict 3D occupancy from 2D Bird's-Eye- View (BEV) features through the Channel-to-Height mechanism, BEV features are insufficient to store all the height information of the scene, which limits performance. This paper proposes LightOcc, an innovative 3D occupancy prediction framework that leverages Lightweight Spatial Embedding to effectively supplement the height clues for the BEV-based representation while maintaining its deployability. Firstly, Global Spatial Sampling is used to obtain the Single-Channel Occupancy from multi-view depth distribution. Spatial-to-Channel mechanism then takes the arbitrary spatial dimension of Single-Channel Occupancy as the feature dimension and extracts Tri-Perspective Views (TPV) Embeddings by 2D convolution. Finally, TPV Embeddings will interact with each other by Lightweight TPV Interaction module to obtain the Spatial Embedding that is optimal supplementary to BEV features. Sufficient experimental results show that LightOcc significantly increases the prediction accuracy of the baseline and achieves stateof- the-art performance on the Occ3D-nuScenes benchmark. Copyright © 2024, The Authors. All rights reserved.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

Pedestrian attribute recognition as label-balanced multi-label learning 24

Pedestrian attribute recognition as label-balanced multi-lab...

引用

Proceedings of the 41st International Conference on Machine Learning

作者： Yibo Zhou Hai-Miao Hu Yirong Xiang Xiaokang Zhang Haotian Wu Beijing Key Laboratory of Digital Media School of Computer Science and Engineering and State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing China Beijing Key Laboratory of Digital Media School of Computer Science and Engineering and State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing China and Hangzhou Innovation Institute of Beihang University Hangzhou China The University of Manchester UK

Rooting in the scarcity of most attributes, realistic pedestrian attribute datasets exhibit unduly skewed data distribution, from which two types of model failures are delivered: (1) label imbalance: model predictions lean greatly towards the side of majority labels; (2) semantics imbalance: model is easily overfitted on the under-represented attributes due to their insufficient semantic diversity. To render perfect label balancing, we propose a novel framework that successfully decouples label-balanced data re-sampling from the curse of attributes co-occurrence, i.e., we equalize the sampling prior of an attribute while not biasing that of the co-occurred others. To diversify the attributes semantics and mitigate the feature noise, we propose a Bayesian feature augmentation method to introduce true in-distribution novelty. Handling both imbalances jointly, our work achieves best accuracy on various popular benchmarks, and importantly, with minimal computational budget.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Teleoperation Based on Hybrid Mapping Method of Human-Robot Heterogeneous Workspace 18th

Teleoperation Based on Hybrid Mapping Method of Human-Robo...

引用

18th Chinese Intelligent systems Conference, CISC 2022

作者： Ye, Shuangxiu Zhang, Ning Zhao, Yongjia State Key Laboratory of Virtual Reality Technology and System Beihang University Beijing100191 China Jiangxi Research Institute Beihang University Jiangxi330000 China

ISBN: (纸本)9789811962028

In this paper, we designed a robot teleoperation system of human-robot heterogeneous workspace which is based on a hybrid mapping method, a combination of terminal pose mapping and joint angle mapping. By using this hybrid mapping method, the problem of choosing the solution that best fits the expectation of the inverse kinematics of the robot is solved, which is difficult to be realized by the simple terminal position mapping. This method uses the RGB-D camera to identify the operator’s joint position information which has little physical constraints on the human body. We also realize the operator’s gesture recognition based on motion-sensing technology which can control the gripper via opening or closing the palm. Some experiments are designed to verify the fine effect of hybrid mapping method and the feasibility of teleoperation system. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Mapping

来源：评论

学校读者我要写书评

暂无评论

Learning Discriminative Representations for Skeleton Based Action Recognition

Learning Discriminative Representations for Skeleton Based A...

引用

Conference on Computer Vision and Pattern Recognition (CVPR)

作者： Huanyu Zhou Qingjie Liu Yunhong Wang State Key Laboratory of Virtual Reality Technology and Systems Beihang University Beijing China Zhongguancun Laboratory Hangzhou Innovation Institute of Beihang University

Human action recognition aims at classifying the category of human action from a segment of a video. Recently, people have dived into designing GCN-based models to extract features from skeletons for performing this task, because skeleton representations are much more efficient and robust than other modalities such as RGB frames. However, when employing the skeleton data, some important clues like related items are also discarded. It results in some ambiguous actions that are hard to be distinguished and tend to be misclassified. To alleviate this problem, we propose an auxiliary feature refinement head (FR Head), which consists of spatial-temporal decoupling and contrastive feature refinement, to obtain discriminative representations of skeletons. Ambiguous samples are dynamically discovered and calibrated in the feature space. Furthermore, FR Head could be imposed on different stages of GCNs to build a multi-level refinement for stronger supervision. Extensive experiments are conducted on NTU RGB+D, NTU RGB+D 120, and NW-UCLA datasets. Our proposed models obtain competitive results from state-of-the-art methods and can help to discriminate those ambiguous samples. Codes are available at https://***/zhysora/FR-Head.

关键词：

来源：评论

学校读者我要写书评

暂无评论

E2NeRF: Event Enhanced Neural Radiance Fields from Blurry Images

E2NeRF: Event Enhanced Neural Radiance Fields from Blurry Im...

引用

International Conference on Computer Vision (ICCV)

作者： Yunshan Qi Lin Zhu Yu Zhang Jia Li State Key Laboratory of Virtual Reality Technology and Systems SCSE Beihang University Beijing Institute of Technology SenseTime and Tetras.AI Peng Cheng Laboratory

Neural Radiance Fields (NeRF) achieves impressive rendering performance by learning volumetric 3D representation from several images of different views. However, it is difficult to reconstruct a sharp NeRF from blurry input as often occurred in the wild. To solve this problem, we propose a novel Event-Enhanced NeRF (E 2 NeRF) by utilizing the combination data of a bio-inspired event camera and a standard RGB camera. To effectively introduce event stream into the learning process of neural volumetric representation, we propose a blur rendering loss and an event rendering loss, which guide the network via modelling real blur process and event generation process, respectively. Moreover, a camera pose estimation framework for real-world data is built with the guidance of event stream to generalize the method to practical applications. In contrast to previous image-based or event-based NeRF, our framework effectively utilizes the internal relationship between events and images. As a result, E 2 NeRF not only achieves image deblurring but also achieves high-quality novel view image generation. Extensive experiments on both synthetic data and real-world data demonstrate that E 2 NeRF can effectively learn a sharp NeRF from blurry images, especially in complex and low-light scenes. Our code and datasets are publicly available at https://***/iCVTEAM/E2NeRF.

关键词：

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：