检索结果-内蒙古大学图书馆

FPSMix: data augmentation strategy for point cloud classification

Frontiers of Computer Science 2025年第2期19卷 105-113页

作者： Taiyan CHEN Xianghua YING Key Lab of Machine Perception(MoE) School of Intelligence Science and TechnologyPeking UniversityBeijing 100871China

Data augmentation is a widely used regularization strategy in deep neural networks to mitigate overfitting and enhance *** the context of point cloud data,mixing two samples to generate new training examples has proven to be *** this paper,we propose a novel and effective approach called Farthest Point Sampling Mix(FPSMix)for augmenting point cloud *** method leverages farthest point sampling,a technique used in point cloud processing,to generate new samples by mixing points from two original point *** key innovation of our approach is the introduction of a significance-based loss function,which assigns weights to the soft labels of the mixed samples based on the classification loss of each part of the new sample that is separated from the two original point *** way,our method takes into account the importance of different parts of the mixed sample during the training process,allowing the model to learn better global *** results demonstrate that our FPSMix,combined with the significance-based loss function,improves the classification accuracy of point cloud models and achieves comparable performance with state-of-the-art data augmentation ***,our approach is complementary to techniques that focus on local features,and their combined use further enhances the classification accuracy of the baseline model.

关键词： point cloud classification data augmentation loss function point cloud understanding point cloud analysis

来源：评论

学校读者我要写书评

暂无评论

Approximate Poisson disk sampling on mesh

引用

Science China(Information Sciences) 2013年第9期56卷 194-205页

作者： GENG Bo ZHANG HuiJuan WANG Heng WANG GuoPing Graphics and Interactive Technology Lab of Dept. of Computer Science Peking University The Key Lab of Machine perception and intelligent MOE

Poisson disk sampling has been widely used in many applications such as remeshing, procedural texturing, object distribution, illumination, etc. While 2D Poisson disk sampling is intensively studied in recent years, direct Poisson disk sampling on 2-manifold surface is rarely covered. In this paper, we present a novel framework which generates approximate Poisson disk distribution directly on mesh, a discrete representation of 2-manifold surfaces. Our framework is easy to implement and provides extra flexibility to specified sampling issues like feature-preserving sampling and adaptive sampling. We integrate the tensor voting method into feature detection and adaptive sample radius calculation. Remeshing as a special downstream application is also addressed. According to our experiment results, our framework is efficient, robust, and widely applicable.

关键词： Poisson disk sampling feature preserving sampling adaptive sampling remeshing tensor voting

来源：评论

学校读者我要写书评

暂无评论

A Forward Masking Auditory Model and Its Application in Speaker Identification and Speech Recognition

引用

中国电子杂志（英文版） 2001年第2期10卷 196-199页

作者： LIU Zhimin WU Xihong ZHEN Bin CHI Huisheng National Key Lab on Machine Perception Center for Information Science Peking University Beijing 100871 China

A novel computational auditory model which simulates the forward-masking mechanism of auditory nerve discharge is presented. Both features based on the model are extracted: FMFRC (forward masking firing-rate cepstrum) and FMSRC (forward masking synchronized rate cepstrum).Isolated-word speech recognition and text-dependent speaker identification experiments based on TI46 are performed. The results show that the new features based on the forward masking model is far more robust than MFCC (mei-frequency cepstrum coefficients) and the performance will be improved compared to the features without such dynamic property. Moreover, the model and the feature extraction method based on it are feasible in practice and promising in robust speech recognition and speaker identification.

关键词： Forward masking MFCC Firingrate cepstrum Synchronized rate cepstrum.

来源：评论

学校读者我要写书评

暂无评论

Newton design:designing CNNs with the family of Newton's methods

引用

Science China(Information Sciences) 2023年第6期66卷 122-137页

作者： Zhengyang SHEN Yibo YANG Qi SHE Changhu WANG Jinwen MA Zhouchen LIN School of Mathematical Sciences Peking University Bytedance AI Lab JD Explore Academy Key Laboratory of Machine Perception School of Intelligence Science and TechnologyPeking University Pazhou Lab

Nowadays, convolutional neural networks(CNNs)have led the developments of machine ***, most CNN architectures are obtained by manual design, which is empirical, time-consuming, and non-transparent. In this paper, we aim at offering better insight into CNN models from the perspective of optimization theory. We propose a unified framework for understanding and designing CNN architectures with the family of Newton's methods, which is referred to as Newton design. Specifically, we observe that the standard feedforward CNN model(PlainNet)solves an optimization problem via a kind of quasi-Newton method. Interestingly, residual network(ResNet)can also be derived if we use a more general quasi-Newton method to solve this problem. Based on the above observations, we solve this problem via a better method,the Newton-conjugate-gradient(Newton-CG)method, which inspires Newton-CGNet. In the network design,we translate binary-value terms in the optimization schemes to dropout layers, so dropout modules naturally appear in the derived CNN structures with specific locations, rather than being an empirical training *** experiments on image classification and text categorization tasks verify that Newton-CGNets perform very competitively. Particularly, Newton-CGNets surpass their counterparts ResNets by over 4% on CIFAR-10 and over 10% on CIFAR-100, respectively.

关键词： CNN dropout optimization method network design Newton's method

来源：评论

学校读者我要写书评

暂无评论

ViWoSG:A distributed scene graph of ultra-massive distributed virtual environments

引用

Science in China(Series F) 2009年第3期52卷 457-469页

作者： WANG GuoPing, LI Sheng, WANG ShaoRong, LU Bin & LI WenHang key lab of machine perception and Intelligent, MOE Department of Computer Science, Peking University, Beijing 100871, China Key Lab of Machine Perception and Intelligent MOE Beijing China Department of Computer Science Peking University Beijing 100871 China

An ultra-massive distributed virtual environment generally consists of ultra-massive terrain data and a large quantity of objects and their attribute data, such as 2D/3D geometric models, audio/video, images, vectors, characteristics, etc. In this paper, we propose a novel method for constructing distributed scene graphs with high extensibility. This method can support high concurrent interaction of clients and implement various tasks such as editing, querying, accessing and motion controlling. Some application experiments are performed to demonstrate its efficiency and soundness.

关键词： VR distributed system scene graph ultra-massive distributed scene graph

来源：评论

学校读者我要写书评

暂无评论

Vehicle Trajectory Collection Using On-Board Multi-Lidars for Driving Behavior Analysis

Vehicle Trajectory Collection Using On-Board Multi-Lidars fo...

引用

7th International Conference on Intelligent Systems and Knowledge Engineering, ISKE 2012

作者： Zhao, Huijing Wang, Chao Yao, Wen Cui, Jinshi Zha, Hongbin Key Lab of Machine Perception Peking University Beijing 100871 China

In order to study the driving behaviors, such as lane change and overtaking, which concerns the relationship between ego and all-surrounding vehicles, it is of great demand in developing an automated system to collect the synchronized motion trajectories that characterize the full course of driving maneuvers in real-world traffic scene. This research proposes a measurement and data processing system, where multiple 2D-Lidars are mounted on a vehicle platform to generate an omni-directional horizontal coverage to the ego-vehicle's surrounding;focusing the driving scenario on motorway, two processing approaches in online and offline procedures are studied for vehicle trajectory extraction by fusing the multi-Lidar data that are acquired during on-road driving. A case study is conducted using a data set collected during 10 min' driving and lasted for 4.1 km long. The performance of trajectory extraction in online and offline procedures is comparatively examined. In addition, a reference vehicle participated in data collection too. The trajectory is analyzed to study its potential in characterizing the situations during vehicle maneuvers. © Springer-Verlag Berlin Heidelberg 2014.

关键词： Vehicles

来源：评论

学校读者我要写书评

暂无评论

Efficient detection of projected concentric circles using four intersection points on a secant line

Efficient detection of projected concentric circles using fo...

引用

作者： Ying, Xianghua Zha, Hongbin Key Lab of Machine Perception Ministry of Education Peking University Beijing China

ISBN: (纸本)9781424421756

Concentric circles are often used for calibration. Based on the geometric properties of concentric circles, we proved that only four intersection edge points on one secant line of the two images of the concentric circles (ICC) are sufficient to determine the whole parameters of the ICC, when the ratio of the radii of two concentric circles is given. Experimental results validate the proposed approach. © 2008 IEEE.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Motion planning for humanoid robots in complex environments based on stance sequence and ZMP reference

Motion planning for humanoid robots in complex environments ...

引用

International Conference on Mechanical Science and Engineering, ICMSE 2012

作者： Liu, Hong Sun, Qing Key Lab of Machine Perception and Intelligence Shenzhen Graduate School Peking University China

ISBN: (纸本)9783037854723

It is a great challenge to plan motion for humanoid robots in complex environments especially when the terrain is cluttered and discrete. To address this problem, a novel method is proposed in this paper by planning the gait according to the stance sequence and ZMP (Zero Moment Point) reference. It consists of two components: an adaptive footstep planner and a walking pattern generator. The adaptive footstep planner can generate the stance path according to the walking rules and adjust the orientation of body relevantly. As the footstep locations are determined, Linear Inverted Pendulum Model (LIPM) is used to generate the walking pattern with a moving ZMP reference. As demonstrated in experiments on the humanoid robot HOAP-2, our method can successfully plan footstep trajectories as well as generate the stable and natural-looking gait in typical cluttered and discrete environments. © (2012) Trans Tech Publications, Switzerland.

关键词： Anthropomorphic robots

来源：评论

学校读者我要写书评

暂无评论

Collision probability based safe path planning for mobile robots in changing environments

Collision probability based safe path planning for mobile ro...

引用

International Conference on Mechanical Science and Engineering, ICMSE 2012

作者： Liu, Hong Wang, Chuangqi Key Lab of Machine Perception and Intelligence Shenzhen Graduate School Peking University China

ISBN: (纸本)9783037854723

Automatic path planning has many applications in robotics, computer-aided design(CAD) and industrial manipulation. The property of safety is vital but seldom taken into consideration by typical path planning. In this paper, collision probability is introduced as an evaluation of crowd degree of environments to get a safer path. The smaller collision probability a node has, the more possibly the node can be extended. Meanwhile, the in/out degree of a node is limited to prevent some nodes to be extended excessively. Through evaluating collision probability on-line, a safe path planning based on DRRTs, called Safe-DRRT, is proposed to provide a path not only feasible but also safe. Finally, a path planner is implemented with Safe-DRRT as a guidance and a local planner. In plentifully crowded experiments with moving obstacles, the proposed method has demonstrated to be competent compared to the state of the art. © (2012) Trans Tech Publications, Switzerland.

关键词： Manipulators

来源：评论

学校读者我要写书评

暂无评论

Reducing Vibration of A Rotating machine with Deep Reinforcement Learning 17

Reducing Vibration of A Rotating Machine with Deep Reinforce...

引用

17th IEEE International Conference on Mechatronics and Automation, ICMA 2020

作者： Tao, Zhang Yian, Deng Fan, Hu Xiangqi, Zhang Yaoyao, Wei Tianlin, Liu Bolchover, Paul Tao, Yu Guishui, Zheng Rongbing, Chen Ping, Zhang Qing, Liu Dingsheng, Luo Peking University Key Lab of Machine Perception Department of Machine Intelligence Beijing100871 China Schlumberger United States

ISBN: (纸本)9781728164151

Reducing vibration of a high-speed rotating machine is an important task as the vibration can damage equipment, increase costs and reduce quality. Faced with the problems, in this paper, we proposed a control tool with several pads and sensors to actively dampening the vibration of the machine. Using Deep Reinforcement Learning (RL) to learn a policy that determines whether to stretch out a pad and the speed of the pads according to the information collected by sensors, our approach can change the displacement of the rotating machine in space by generating an opposite force with a pad while it touches the wall so as to reduce the vibration. We evaluate our approach against two baselines, a control method with all the pads stretched out and a method without equipping the control tool. Our results indicate that simple RL algorithms in controlling the pad tool have effectiveness in reducing the vibration of the machine. © 2020 IEEE.

关键词： Reinforcement learning

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：