tp钱包最新版下载|dbl
如何掌握LabVIEW的数值数据类型-I8、I16、I32、SGL、DBL等 - 墨天轮
如何掌握LabVIEW的数值数据类型-I8、I16、I32、SGL、DBL等 - 墨天轮
排行 数据库百科 核心案例 行业报告 月度解读 大事记 产业图谱
中国数据库
向量数据库
时序数据库
实时数据库
搜索引擎
空间数据库
图数据库
数据仓库
大调查
2021年报告
2022年报告
年度数据库
2020年openGauss
2021年TiDB
2022年PolarDB
2023年OceanBase
首页
资讯
数说
活动
大会
学习 课程中心 推荐优质内容、热门课程学习路径 预设学习计划、达成学习目标知识图谱 综合了解技术体系知识点课程库 快速筛选、搜索相关课程视频学习 专业视频分享技术知识电子文档 快速搜索阅览技术文档文档
问答
工具 SQLRUN 在线数据库即时SQL运行平台数据库在线实训平台 实操环境、开箱即用、一键连接AWR分析 上传AWR报告,查看分析结果SQL格式化 快速格式化绝大多数SQL语句SQL审核 审核编写规范,提升执行效率PLSQL解密 解密超4000字符的PL/SQL语句OraC函数 查询Oracle C 函数的详细描述智能助手小墨 关于数据库相关的问题,您都可以问我 数据库百科核心案例行业报告月度解读大事记产业图谱我的订单
登录后可立即获得以下权益 免费培训课程 收藏优质文章 疑难问题解答 下载专业文档 签到免费抽奖 提升成长等级 立即登录 登录 注册 登录 注册 首页资讯数说活动大会课程文档排行问答我的订单 微信扫码 复制链接 新浪微博 分享数说 采集到收藏夹 分享到数说 首页 /
如何掌握LabVIEW的数值数据类型-I8、I16、I32、SGL、DBL等 如何掌握LabVIEW的数值数据类型-I8、I16、I32、SGL、DBL等 虚拟仪器技术及应用 2022-04-20 12017
所有人应该都知道,可用于加减乘除等数学运算的数字(如5,23,12.6,13.7)都应该用数值数据类型来表示。但是大部分人对数值数据类型的分类认识仅有整型、单精度和双精度三个分类,对使用的认识也仅停留在整数选用整型,小数选用单精度或双精度这种很粗的层次。而LabVIEW的数值数据类型实在有点多(有I8、I16、I32、U8、U16、U32、SGL、DBL、EXT、CSG、CDB、CXT共12种),且在使用时是通过设置的方式来指定数据类型(而不是文本式语言直观的文字表示方式,如int a或者double a),很多人就有点蒙了。那么,应该如何掌握LabVIEW的数值数据类型呢?第一步是要先认识LabVIEW的所有数值数据类型。(一)LabVIEW的数值数据类型介绍LabVIEW的数值型控件都会默认一种类型,且不同的数据类型的端口图标会有不同的外观。下图展示了所有LabVIEW的数值数据类型的数据端口外观。数一下图标数量,发现足足有24种,是不是有点崩溃?但不要着急,大家先注意一下各个图标上的文字,会发现,有两两一对的图标上标的文字是一样的,如下图:这两两一对被框起来的图标其实是同一种数据类型,他们只是外观不同,所以虽然有24种图标,其实只有12种数据类型。且通过这个介绍,大家也认识到,可以通过图标上的文字来识别数据类型。根据上图中各个图标上的文字,可以知道LabVIEW共有I8、I16、I32、U8、U16、U32、SGL、DBL、EXT、CSG、CDB、CXT共12种数值数据类型。这种外观不同但是属于同种类型的数据端口外观可以通过右键点击数据端口,在弹出的菜单中勾选和不勾选“View As Icon”菜单项来实现切换,如下图所示。勾选“View As Icon”显示的是比较大的图标外观,不勾选“View As Icon”显示的较小的端口外观。下面统一以端口外观的形式继续介绍LabVIEW的数值数据类型,如下图。根据我们使用的数据类型,上面的12种图标又可以分为三类,分别对应整数(如5,34,125)、小数(如0.12,1.34,567.8)和复数(如2+3i,5+6.2i,1.3+4.5i),如下图:大家此时也可能注意到,端口的颜色也不一样。颜色确实是LabVIEW用于表达数据类型不同的一个手段。上图中,蓝色用于表示整数,橙色用于表示小数(对复数,表示复数的实部和虚部为小数)。我们接下来先来看整数。整数进一步分为6种数据类型,分别为I8、I16、I32、U8、U16、U32。整数的这种分类是根据数据范围来划分的,其中I8可表示的数据范围是-128~127,I16表示的数据范围是-32768~32767,I32表示的数据范围是-2147484648~2147483637,U8表示的数据范围是0~255,U16表示的数据范围是0~65535,U32表示的数据范围是0~4294967925。大家还可以从各种整数数据类型的名字来进一步了解。如对I8,I表示Integer(整数的英文),8表示用8位二进制表示一个数,那么I8表示的是8位有符号整数(带正负号的整数)。同理,I16表示的是16位的有符号整数,I32表示的是32位的有符号整数。位数越多,那么可表示的数据范围越大,所以大家也不必要把各种数据类型表示的数据范围背诵下来,大体对各种数据类型表示的数据范围有印象,且知道表示的数据类型的范围从小到大的顺序为 I8、I16、I32就可以。同样,对U8,U表示Unsigned Integer(无符号整数的英文),8表示用8位二进制表示一个数,其不表示符号位,所以其表示的整数范围与I8不一样,I8为-128~127,U8为0~255。继续来看小数。小数进一步分为3种数据类型,分别为SGL、DBL、EXT。同样,小数的这种分类也是根据数据范围来细分的,其中SGL可表示的数据范围为:最小正数1.40e-45,最大正数3.40e+38,最小负数-1.40e-45,最大负数-3.40e+38;DBL可表示的数据范围为:最小正数4.94e-324,最大正数1.79e+308,最小负数-4.94e-324,最大负数-1.79e+308;EXT可表示的数据范围为:最小正数6.48e-4966,最大正数1.19e+4932,最小负数-6.48e-4966,最大负数-1.19e+4932。大家同样可以从各种小数数据类型的名字来进一步了解。如对SGL,英文全称为Single,表示单精度浮点数,跟其他语言的单精度浮点数数据类型是一样的;如对DBL,英文全称为Double,表示双精度浮点数,跟其他语言的双精度浮点数的数据类型也是一样的;而EXT,英文全称为Extended,表示扩展精度浮点数,很多其他语言都没有这种数据类型,这是因为LabVIEW是专门用于虚拟仪器的语言,很多时候对数据有很大的范围和精度要求。继续来看复数。复数也进一步细分为3种数据类型,分别为CSG、CDB和CXT。其中,CSG表示实部和虚部用SGL表示的复数,CSG中,C为Complex,表示复数;SG表示SGL,Single,表示实部和虚部用单精度数据类型表示。CDB表示实部和虚部用DBL表示的复数,CDB中,C为Complex,表示复数;DB表示DBL,Double,表示实部和虚部用双精度数据类型表示。CXT表示实部和虚部用EXT表示的复数,CXT中,C为Complex,表示复数;XT表示EXT,Extended,表示实部和虚部用扩展精度数据类型表示。好了,到此大家应该全面了解了LabVIEW的所有数值数据类型。下来就是如何选用LabVIEW的数值数据类型的问题。(二)LabVIEW的数值数据类型选用如果充分了解了LabVIEW的数值数据类型,对其选用是极为简单的。大家其实也注意到了不同的数据类型本质是其表示的数据范围和精度不一样,那么就根据你要使用的场合、计算精度、范围和存储空间要求选择即可。具体方法为:(1)根据要使用的数据是整数、小数还是复数确定数值数据类型大的分类选择;(2)根据要使用的数据的数据范围选择一个可包含其范围的数据类型。比如,要表示的数据是整数,其范围为138-380,则应选择的最合适的数据类型为U16或者I16。当然,I32、U32、SGL、DBL、EXT也是可以满足使用要求的,就是会浪费存储空间。又比如,要表示的数据是小数,其范围为-1245.9-+9999.9,那么应选择的最合适的数据类型为SGL。当然,DBL和EXT也是可以满足使用要求的,就是会浪费存储空间。下面整理出了各种数据类型的数值范围,需要选择数据类型时,大家可以参考该表格。(三)练习题若要表示的数据范围分别为-126-125和-50000000000-50000000000,应分别选用哪种数据类型最合适?若要表示的数据是3+4i,又应该选用哪种数据类型最合适?
labviewi8 文章转载自虚拟仪器技术及应用,如果涉嫌侵权,请发送邮件至:contact@modb.pro进行举报,并提供相关证据,一经查实,墨天轮将立刻删除相关内容。 评论 领墨值 有奖问卷 意见反馈 客服小墨
5-临床数据管理员(DM)需知道的英文表达或缩写(持续更新) - 知乎
5-临床数据管理员(DM)需知道的英文表达或缩写(持续更新) - 知乎首发于DM入门切换模式写文章登录/注册5-临床数据管理员(DM)需知道的英文表达或缩写(持续更新)DM潘小白临床数据管理员,DM。B站同名。一、机构或组织CRO(Contract Research Organization)合同研究组织SMO(Site Management Organization) 现场管理组织Sponsor:申办方FDA(Food and Drug Administration)美国食品药品监督管理局NMPA(National Medical Products Administration)国家药品监督管理局CDE(Center for Drug Evaluation)国家药品监督管理局药品审批中心EC(Ethics Committee)/IEC(Independent Ethics Committee)(独立)伦理委员会IRB(Institutional Review Board)机构审查委员会伦理委员会和机构审查委员会是一个意思,EC是欧盟的称呼(摘自Wikipedia:The ethics committee is an independent body in a member state of the European Union。),而IRB是美国的称呼(摘自Wikipedia:"IRB" is a generic term used in the United States by the FDA and HHS)。ICH(International Council for Harmonization)国际协调会议CDISC(Clinical Data Interchange Standards Consortium)临床数据交换标准协会DSMB/IDMC(Data Safety and Monitoring Board/Independent Data Monitoring Committee)数据与安全监察委员会/独立数据监察委员会ISO(International Standard Organization)国际化标准组织WHO(World Health Organization)世界卫生组织SCDM(Society for Clinical Data Management)美国临床试验数据管理学会CDMC(Clinical Data Management Working Group of China)中国临床试验数据管理学组DIA(Drug Information Association)药物信息协会二、岗位或职位DM(Data Manager)数据管理员CRA(Clinical Research Assistant)临床监查员CRC(Clinical Research Coordinator)临床协调员CTA(Clinical Trial Assistant)临床试验助理PI(Principal Investigator)主要研究者Co-I(Co-Investigator)共同研究者,是ICH-GCP之前的一个术语,后来被Sub-I取代。(Co-investigator is a term that pre-dates the ICH GCP's. It has subsequently been replaced by the term sub-investigator.)可以执行全部或部分PI功能,但他们不承担研究的主要责任,受PI的监督,协助其进行研究项目的管理和领导。CI(Coordinating Investigator)协调研究者An investigator assigned the responsibility for the coordination of investigators at different centres participating in a multicentre trial. (ICH-GCP)被指定负责协调参加一项多中心试验的各中心研究者工作的一名研究者。Sub-I/SI(Sub-Investigator)次要研究者SP(SAS Programmer)SAS程序员Biostatistician:生物统计师DBD(Database Designer)数据库设计员/建库员PV(Pharmacovigilance)药物警戒MSL(Medical Science Liaison)医学联络官MA(Medical Advisor)医学顾问BD(Business Development)商务拓展PM(Project Manager)项目经理 APM(Assistant Project Manager) 项目经理助理三、文件GCP(Good Clinical Practice)临床试验质量管理规范GCDMP(Good Clinical Data Management Practice)临床数据质量管理规范ICF (Informed Consent Form)知情同意书IB(Investigator's Brochure)研究者手册SOP(Standard Operation Procedure)标准操作规程TMF(Trial Master File)临床试验中央文件夹CRF(Case Report form)病例报告表eCRF(Electronic Case Report form)电子病例报告表Mock CRF:模拟CRF。未完成建库前设计的用于递交伦理的CRF,简称CRF。eCCG(eCRF Completion Guide)eCRF填写指南DMP(Data Management Plan)数据管理计划DVP(Data Validation Plan)数据核查计划DVR(Data Validation Report)数据核查报告DMR(Data Management Report)数据管理报告SAP(Statistical Analysis Plan)统计分析计划DCF(Data Clarification Form)数据澄清表Protocol:方案四、工作相关EDC(Electronic Data Capture)电子数据采集CTMS(Clinical Trial Management System)临床试验信息管理系统CDMS(Clinical Data Management System)临床数据管理系统以上三者系统的区别与联系:FPFV(First Patient First Visit)第一例患者的第一次访视LPLV(Last Patient Last Visit)最后一例患者最后一次访视SDV(Source Data Verification)原始数据核查PD(Protocol Deviation)方案偏离EOT(End of Treatment)中止试验DBL(Database Locking)数据库锁定UAT(User Acceptance Testing)用户接受测试QC(Quality Control)质量控制ADR(Adverse Drug Reaction)药品不良反应AE(Adverse Event)不良事件SAE(Serious Adverse Event)严重不良事件SUSAR(Suspected Unexpected Serious Adverse Reaction )可疑的非预期的严重不良反应site:中心subject:受试者visit:访视screen:筛选enrollment:入组query:质疑timeline:时间表CRF design:CRF设计database setup:数据库建立system validation and change control:系统验证及变更控制data validation programming/edit check:逻辑核查data entry:数据录入coding:编码handling lab normal ranges:实验室指标正常值范围SAE reconciliation:SAE一致性检验data audit:数据稽查database lock/unlock:数据库锁定及解锁document management:文档管理archive:归档training:培训transfer/extraction of data:数据转入转出external data/non-CRF data:外部数据access control:权限控制electronic signature:电子签名blind review:盲态审核audit trail:稽查轨迹protocol deviation:方案偏离新增(重要级别与上面相比较低):ICSR(Individual Case Safety Reports)个例药品不良反应报告PMDA(Pharmaceutical and Medical Devices Agency)日本药品及医疗器械管理局MAH(Marketing Authorization Holder)上市许可持有人编辑于 2021-06-08 20:38临床试验临床数据管理赞同 723 条评论分享喜欢收藏申请转载文章被以下专栏收录DM入门介绍临床数据管理相关的基础知识,适合无经
全网呕血整理:关于YOLO v3原理分析 - 知乎
全网呕血整理:关于YOLO v3原理分析 - 知乎首发于程序员之家切换模式写文章登录/注册全网呕血整理:关于YOLO v3原理分析华为云开发者联盟已认证账号摘要:YOLO系列的目标检测算法可以说是目标检测史上的宏篇巨作,接下来我们来详细介绍一下YOLO v3算法内容。算法基本思想首先通过特征提取网络对输入特征提取特征,得到特定大小的特征图输出。输入图像分成13×13的grid cell,接着如果真实框中某个object的中心坐标落在某个grid cell中,那么就由该grid cell来预测该object。每个object有固定数量的bounding box,YOLO v3中有三个bounding box,使用逻辑回归确定用来预测的回归框。网络结构上图DBL是Yolo v3的基本组件。Darknet的卷积层后接BatchNormalization(BN)和LeakyReLU。除最后一层卷积层外,在yolo v3中BN和LeakyReLU已经是卷积层不可分离的部分了,共同构成了最小组件。主干网络中使用了5个resn结构。n代表数字,有res1,res2, … ,res8等等,表示这个res_block里含有n个res_unit,这是Yolo v3的大组件。从Yolo v3开始借鉴了ResNet的残差结构,使用这种结构可以让网络结构更深。对于res_block的解释,可以在上图网络结果的右下角直观看到,其基本组件也是DBL。在预测支路上有张量拼接(concat)操作。其实现方法是将darknet中间层和中间层后某一层的上采样进行拼接。值得注意的是,张量拼接和Res_unit结构的add的操作是不一样的,张量拼接会扩充张量的维度,而add只是直接相加不会导致张量维度的改变。Yolo_body一共有252层。23个Res_unit对应23个add层。BN层和LeakyReLU层数量都是72层,在网络结构中的表现为:每一层BN后面都会接一层LeakyReLU。上采样和张量拼接操作各2个,5个零填充对应5个res_block。卷积层一共有75层,其中有72层后面都会接BatchNormalization和LeakyReLU构成的DBL。三个不同尺度的输出对应三个卷积层,最后的卷积层的卷积核个数是255,针对COCO数据集的80类:3×(80+4+1)=255,3表示一个grid cell包含3个bounding box,4表示框的4个坐标信息,1表示置信度。下图为具体网络结果图。输入映射到输出不考虑神经网络结构细节的话,总的来说,对于一个输入图像,YOLO3将其映射到3个尺度的输出张量,代表图像各个位置存在各种对象的概率。我们看一下YOLO3共进行了多少个预测。对于一个416*416的输入图像,在每个尺度的特征图的每个网格设置3个先验框,总共有 13*13*3 + 26*26*3 + 52*52*3 = 10647 个预测。每一个预测是一个(4+1+80)=85维向量,这个85维向量包含边框坐标(4个数值),边框置信度(1个数值),对象类别的概率(对于COCO数据集,有80种对象)。边界框预测(Bounding Box Prediction)Yolo v3关于bounding box的初始尺寸还是采用Yolo v2中的k-means聚类的方式来做,这种先验知识对于bounding box的初始化帮助还是很大的,毕竟过多的bounding box虽然对于效果来说有保障,但是对于算法速度影响还是比较大的。在COCO数据集上,9个聚类如下表所示,注这里需要说明:特征图越大,感受野越小。对小目标越敏感,所以选用小的anchor box。特征图越小,感受野越大。对大目标越敏感,所以选用大的anchor box。Yolo v3采用直接预测相对位置的方法。预测出b-box中心点相对于网格单元左上角的相对坐标。直接预测出(tx,ty,tw,th,t0),然后通过以下坐标偏移公式计算得到b-box的位置大小和confidence。tx、ty、tw、th就是模型的预测输出。cx和cy表示grid cell的坐标,比如某层的feature map大小是13×13,那么grid cell就有13×13个,第0行第1列的grid cell的坐标cx就是0,cy就是1。pw和ph表示预测前bounding box的size。bx、by、bw和bh就是预测得到的bounding box的中心的坐标和size。在训练这几个坐标值的时候采用了sum of squared error loss(平方和距离误差损失),因为这种方式的误差可以很快的计算出来。注:这里confidence = Pr(Object)*IoU 表示框中含有object的置信度和这个box预测的有多准。也就是说,如果这个框对应的是背景,那么这个值应该是 0,如果这个框对应的是前景,那么这个值应该是与对应前景 GT的IoU。Yolo v3使用逻辑回归预测每个边界框的分数。如果边界框与真实框的重叠度比之前的任何其他边界框都要好,则该值应该为1。如果边界框不是最好的,但确实与真实对象的重叠超过某个阈值(Yolo v3中这里设定的阈值是0.5),那么就忽略这次预测。Yolo v3只为每个真实对象分配一个边界框,如果边界框与真实对象不吻合,则不会产生坐标或类别预测损失,只会产生物体预测损失。多尺度预测在上面网络结构图中可以看出,Yolo v3设定的是每个网格单元预测3个box,所以每个box需要有(x, y, w, h, confidence)五个基本参数。Yolo v3输出了3个不同尺度的feature map,如上图所示的y1, y2, y3。y1,y2和y3的深度都是255,边长的规律是13:26:52。每个预测任务得到的特征大小都为N ×N ×[3∗(4+1+80)] ,N为格子大小,3为每个格子得到的边界框数量, 4是边界框坐标数量,1是目标预测值,80是类别数量。对于COCO类别而言,有80个类别的概率,所以每个box应该对每个种类都输出一个概率。所以3×(5 + 80) = 255。这个255就是这么来的。Yolo v3用上采样的方法来实现这种多尺度的feature map。在Darknet-53得到的特征图的基础上,经过六个DBL结构和最后一层卷积层得到第一个特征图谱,在这个特征图谱上做第一次预测。Y1支路上,从后向前的倒数第3个卷积层的输出,经过一个DBL结构和一次(2,2)上采样,将上采样特征与第2个Res8结构输出的卷积特征张量连接,经过六个DBL结构和最后一层卷积层得到第二个特征图谱,在这个特征图谱上做第二次预测。Y2支路上,从后向前倒数第3个卷积层的输出,经过一个DBL结构和一次(2,2)上采样,将上采样特征与第1个Res8结构输出的卷积特征张量连接,经过六个DBL结构和最后一层卷积层得到第三个特征图谱,在这个特征图谱上做第三次预测。就整个网络而言,Yolo v3多尺度预测输出的feature map尺寸为y1:(13×13),y2:(26×26),y3:(52×52)。网络接收一张(416×416)的图,经过5个步长为2的卷积来进行降采样(416 / 2ˆ5 = 13,y1输出(13×13)。从y1的倒数第二层的卷积层上采样(x2,up sampling)再与最后一个26×26大小的特征图张量连接,y2输出(26×26)。从y2的倒数第二层的卷积层上采样(x2,up sampling)再与最后一个52×52大小的特征图张量连接,y3输出(52×52)感受一下9种先验框的尺寸,下图中蓝色框为聚类得到的先验框。黄色框式ground truth,红框是对象中心点所在的网格。预测框的3种情况预测框一共分为三种情况:正例(positive)、负例(negative)、忽略样例(ignore)。(1)正例:任取一个ground truth, 与上面计算的10647个框全部计算IOU, IOU最大的预测框, 即为正例。并且一个预测框, 只能分配给一个ground truth。 例如第一个ground truth已经匹配了一个正例检测框, 那么下一个ground truth, 就在余下的10646个检测框中, 寻找IOU最大的检测框作为正例。ground truth的先后顺序可忽略。正例产生置信度loss、检测框loss、类别loss。预测框为对应的ground truth box标签(使用真实的x、y、w、h计算出); 类别标签对应类别为1, 其余为0; 置信度标签为1。(2)忽略样例:正例除外, 与任意一个ground truth的IOU大于阈值(论文中使用5), 则为忽略样例。忽略样例不产生任何loss。为什么会有忽略样例?由于Yolov3采用了多尺度检测, 那么再检测时会有重复检测现象. 比如有一个真实物体,在训练时被分配到的检测框是特征图1的第三个box,IOU达0.98,此时恰好特征图2的第一个box与该ground truth的IOU达0.95,也检测到了该ground truth,如果此时给其置信度强行打0的标签,网络学习效果会不理想。(3)负例:正例除外(与ground truth计算后IOU最大的检测框,但是IOU小于阈值,仍为正例), 与全部ground truth的IOU都小于阈值(0.5), 则为负例。负例只有置信度产生loss, 置信度标签为0。如下图所示:λ为权重参数, 用于控制检测框loss, obj与noobj的置信度loss, 以及类别对于正类而言, 1ijobj输出为1; 对于负例而言, 1ijnoobj输出为1; 对于忽略样例而言, 全部为0;类别采用交叉熵作为损失函数。类别预测类别预测方面Yolo v2网络中的Softmax分类器,认为一个目标只属于一个类别,通过输出Score大小,使得每个框分配到Score最大的一个类别。但在一些复杂场景下,一个目标可能属于多个类(有重叠的类别标签),因此Yolo v3用多个独立的Logistic分类器替代Softmax层解决多标签分类问题,且准确率不会下降。举例说明,原来分类网络中的softmax层都是假设一张图像或一个object只属于一个类别,但是在一些复杂场景下,一个object可能属于多个类,比如你的类别中有woman和person这两个类,那么如果一张图像中有一个woman,那么你检测的结果中类别标签就要同时有woman和person两个类,这就是多标签分类,需要用Logistic分类器来对每个类别做二分类。Logistic分类器主要用到sigmoid函数,该函数可以将输入约束在0到1的范围内,因此当一张图像经过特征提取后的某一类输出经过sigmoid函数约束后如果大于0.5,就表示该边界框负责的目标属于该类。物体分数和类置信度物体分数:表示一个边界框包含一个物体的概率,对于红色框和其周围的框几乎都为1,但边角的框可能几乎都为0。物体分数也通过一个sigmoid函数,表示概率值。类置信度:表示检测到的物体属于一个具体类的概率值,以前的YOLO版本使用softmax将类分数转化为类概率。在YOLOv3中作者决定使用sigmoid函数取代,原因是softmax假设类之间都是互斥的,例如属于“Person”就不能表示属于“Woman”,然而很多情况是这个物体既是“Person”也是“Woman”。输出处理我们的网络生成10647个锚框,而图像中只有一个狗,怎么将10647个框减少为1个呢?首先,我们通过物体分数过滤一些锚框,例如低于阈值(假设0.5)的锚框直接舍去;然后,使用NMS(非极大值抑制)解决多个锚框检测一个物体的问题(例如红色框的3个锚框检测一个框或者连续的cell检测相同的物体,产生冗余),NMS用于去除多个检测框。具体使用以下步骤:抛弃分数低的框(意味着框对于检测一个类信心不大);当多个框重合度高且都检测同一个物体时只选择一个框(NMS)。为了更方便理解,我们选用上面的汽车图像。首先,我们使用阈值进行过滤一部分锚框。模型有19*19*3*85个数,每个盒子由85个数字描述。将(19,19,3,85)分割为下面的形状:box_confidence:(19,19,3,1)表示19*19个cell,每个cell的 3个框,每个框有物体的置信度概率;boxes:(19,19,3,4)表示每个cell 的3个框,每个框的表示;box_class_probs:(19,19,3,80)表示每个cell的3个框,每个框80个类检测概率。每个锚框我们计算下面的元素级乘法并且得到锚框包含一个物体类的概率,如下图:即使通过类分数阈值过滤一部分锚框,还剩下很多重合的框。第二个过程叫NMS,里面有个IoU,如下图所示。实现非极大值抑制,关键在于:选择一个最高分数的框;计算它和其他框的重合度,去除重合度超过IoU阈值的框;回到步骤1迭代直到没有比当前所选框低的框。Loss Function在Yolo v3的论文里没有明确提出所用的损失函数,确切地说,Yolo系列论文里面只有Yolo v1明确提了损失函数的公式。在Yolo v1中使用了一种叫sum-square error的损失计算方法,只是简单的差方相加。我们知道,在目标检测任务里,有几个关键信息是需要确定的:(x,y),(w,h),class,confidence 。根据关键信息的特点可以分为上述四类,损失函数应该由各自特点确定。最后加到一起就可以组成最终的loss function了,也就是一个loss function搞定端到端的训练。yolov3网络硬核讲解(视频)视频地址:https://www.bilibili.com/video/BV12y4y1v7L6?from=search&seid=442233808730191461真实值是如何编码预测锚框的设计锚框与目标框做iou本文分享自华为云社区《YOLOV3 原理分析(全网资料整理)》,原文作者:lutianfei 。点击关注,第一时间了解华为云新鲜技术~发布于 2021-01-18 10:43目标检测ResNet算法赞同 1506 条评论分享喜欢收藏申请转载文章被以下专栏收录程序员之家欢迎投稿
dblp: computer science bibliography
dblp: computer science bibliography
Stop the war!Остановите войну!solidarity - (ua) - (ru)news - (ua) - (ru)donate - donate - donatefor scientists:ERA4UkraineAssistance in GermanyUkrainian Global University#ScienceForUkraine
default search actioncombined dblp searchauthor searchvenue searchpublication searchask othersGoogleGoogle ScholarSemantic ScholarInternet Archive ScholarCiteSeerXORCID
Welcome to dblp
Search dblp see FAQHow to use the dblp search?How to use the dblp search API?Which technology does dblp use for searching?powered by CompleteSearch, courtesy of Hannah Bast, University of Freiburg
Full-text search see FAQHow to use the dblp search?How to use the dblp search API?Which technology does dblp use for searching?
DagstuhlTrier
> Home
Please enter a search query see FAQ for more informationHow to use the dblp search?Which technology does dblp use for searching the website?case-insensitive prefix search: default e.g., sig matches "SIGIR" as well as "signal"exact word search: append dollar sign ($) to worde.g., graph$ matches "graph", but not "graphics"boolean and: separate words by spacee.g., codd modelboolean or: connect words by pipe symbol (|)e.g., graph|networkUpdate May 7, 2017: Please note that we had to disable the phrase search operator (.) and the boolean not operator (-) due to technical problems. For the time being, phrase search queries will yield regular prefix search result, and search terms preceded by a minus will be interpreted as regular (positive) search terms.Author search resultsexport search results asfirst 1000 hits only:XMLJSONJSONPno matches
Venue search resultsexport search results asfirst 1000 hits only:XMLJSONJSONPno matches
Refine listrefine by authorno optionstemporarily not availablerefine by venueno optionstemporarily not availablerefine by typeno optionstemporarily not availablerefine by access no optionstemporarily not availablerefine by yearno optionstemporarily not availablePublication search resultsexport search results asfirst 1000 hits only:XMLJSONJSONPBibTeXfound 7,130,271 matchesskipping 7,130,271 more matches[next >>]loading more results failed to load more results, please try again laterbrowse authors | editorsABCDEFGHIJKLMNOPQRSTUVWXYZbrowse journalsABCDEFGHIJKLMNOPQRSTUVWXYZby publisherbrowse conferences | workshopsABCDEFGHIJKLMNOPQRSTUVWXYZbrowse seriesCoRRLNCSCEUR-WSLNEEIFIPLNIEPTCSLIPICSotherbrowse monographsbooks & thesesreference worksedited collectionsdblp blog
2024-01-01: 7 million publications
[News]
read as PDF
(read full post)
2023-05-22: DTD update May 2023
[News]
(updated 2023-06-28) A few days ago, we discussed the new dataset publications in dblp. As a preparation for more and more detailed datasets we slightly modify the DTD that defines the structure of our XML data export. A quick reminder: you can download the dblp dataset as a single XML […]
(read full post)
2023-04-27: Dataset publications in dblp
[Blog]
[Feature Spotlight]
Datasets and other research artifacts are a major topic in the scientific community in the recent years. Many ongoing projects focus on improving the standardization, publication and citation of these artifacts. Currently, the dblp team is involved in three of them: NFDI4DataScience, NFDIxCS, and Unknown Data. As part of these […]
(read full post)
2022-11-14: Building the German Research Data Infrastructure NFDI – for and with Computer Science
[Press Release]
On November 4, 2022, the Joint Science Conference (GWK) selected Schloss Dagstuhl – Leibniz Center for Informatics and the consortium NFDIxCS for federal and state funding within the German National Research Data Infrastructure (NFDI). The consortium will be funded in the double-digit millions of Euros and over a duration of five […]
(read full post)
2022-09-09: Updates to the dblp RDF schema
[News]
In the six months since the release of the dblp RDF dump and its persistent snapshot releases, the RDF dump has been downloaded a total of about a thousand times. We are pleased to see that the community is interested in using our semantic data in their research and beyond. […]
(read full post)
more blog postsAbout dblpThe dblp computer science bibliography provides open bibliographic information on major computer science journals and proceedings.
Originally created at the University of Trier in 1993, dblp is now operated and further developed by Schloss Dagstuhl. For more information check out our F.A.Q.
dblp statistics# of publications: 7,130,271# of authors: 3,459,863# of conferences: 6,596# of journals: 1,861# of records added to dblp:more statisticsXML dataYou may download the raw dblp data in a single XML file. A simple DTD is available. The paper DBLP - Some Lessons Learned documents technical details of this XML file. In the appendix DBLP XML Requests you may find the description of a primitive dblp query API. All metadata is released as open data under CC0 1.0 license. RSS feedsdblp blognew issues and volumesRelated resourcesACM Digital LibraryIEEE Xplore | CSDLarXiv.orgSemantic ScholarCiteSeerXBibSonomyINSPIRE-HEPMathSciNetPubMedRePEczbMATHmore external linksSocial media links a service of homeblogstatisticsupdate feedXML dumpRDF dumpbrowsepersonsconferencesjournalsseriessearchsearch dblplookup by IDaboutf.a.q.teamlicenseprivacyimprintnfdidblp is part of theGerman National ResearchData Infrastructure (NFDI)NFDI4DataScienceORKGCEURMyBinderevents | twitter | publicationsNFDIxCSevents | twitter
manage site settingsTo protect your privacy, all features that rely on external API calls from your browser are turned off by default. You need to opt-in for them to become active. All settings here will be stored as cookies with your web browser. For more information see our F.A.Q.Unpaywalled article linksAdd open access links from to the list of external document links (if available).load links from unpaywall.orgPrivacy notice: By enabling the option above, your browser will contact the API of unpaywall.org to load hyperlinks to open access articles. Although we do not have any reason to believe that your call will be tracked, we do not have any control over how the remote server uses your data. So please proceed with care and consider checking the Unpaywall privacy policy.Archived links via Wayback MachineFor web page which are no longer available, try to retrieve content from the of the Internet Archive (if available).load content from archive.orgPrivacy notice: By enabling the option above, your browser will contact the API of archive.org to check for archived content of web pages that are no longer available. Although we do not have any reason to believe that your call will be tracked, we do not have any control over how the remote server uses your data. So please proceed with care and consider checking the Internet Archive privacy policy.Reference listsAdd a list of references from , , and to record detail pages.load references from crossref.org and opencitations.netPrivacy notice: By enabling the option above, your browser will contact the APIs of crossref.org, opencitations.net, and semanticscholar.org to load article reference information. Although we do not have any reason to believe that your call will be tracked, we do not have any control over how the remote server uses your data. So please proceed with care and consider checking the Crossref privacy policy and the OpenCitations privacy policy, as well as the AI2 Privacy Policy covering Semantic Scholar.Citation dataAdd a list of citing articles from and to record detail pages.load citations from opencitations.netPrivacy notice: By enabling the option above, your browser will contact the API of opencitations.net and semanticscholar.org to load citation information. Although we do not have any reason to believe that your call will be tracked, we do not have any control over how the remote server uses your data. So please proceed with care and consider checking the OpenCitations privacy policy as well as the AI2 Privacy Policy covering Semantic Scholar.OpenAlex dataLoad additional information about publications from .load data from openalex.orgPrivacy notice: By enabling the option above, your browser will contact the API of openalex.org to load additional information. Although we do not have any reason to believe that your call will be tracked, we do not have any control over how the remote server uses your data. So please proceed with care and consider checking the information given by OpenAlex. retrieved on 2024-03-07 11:05 CET from data curated by the dblp team all metadata released as open data under CC0 1.0 licensesee also: Terms of Use | Privacy Policy | Imprintdblp was originally created in 1993 at: since 2018, dblp has been operated and maintained by: the dblp computer science bibliography is funded and supported by:
一文看懂YOLO v3 - 知乎
一文看懂YOLO v3 - 知乎切换模式写文章登录/注册一文看懂YOLO v3小绿叶人工智能小绿叶我的CSDN博客:https://blog.csdn.net/litt1e我的公众号:工科宅生活论文地址:https://pjreddie.com/media/files/papers/YOLOv3.pdf论文:YOLOv3: An Incremental ImprovementYOLO系列的目标检测算法可以说是目标检测史上的宏篇巨作,接下来我们来详细介绍一下YOLO v3算法内容,v3的算法是在v1和v2的基础上形成的,所以有必要先回忆:一文看懂YOLO v2,一文看懂YOLO v2。网络结构从这儿盗了张图,这张图很好的总结了YOLOV3的结构,让我们对YOLO有更加直观的理解。DBL:代码中的Darknetconv2d_BN_Leaky,是yolo_v3的基本组件。就是卷积+BN+Leaky relu。resn:n代表数字,有res1,res2, … ,res8等等,表示这个res_block里含有多少个res_unit。不懂resnet请戳这儿concat:张量拼接。将darknet中间层和后面的某一层的上采样进行拼接。拼接的操作和残差层add的操作是不一样的,拼接会扩充张量的维度,而add只是直接相加不会导致张量维度的改变。后面我们一起分析网络一些细节与难懂的地方backbone:darknet-53为了达到更好的分类效果,作者自己设计训练了darknet-53。作者在ImageNet上实验发现这个darknet-53,的确很强,相对于ResNet-152和ResNet-101,darknet-53不仅在分类精度上差不多,计算速度还比ResNet-152和ResNet-101强多了,网络层数也比他们少。Yolo_v3使用了darknet-53的前面的52层(没有全连接层),yolo_v3这个网络是一个全卷积网络,大量使用残差的跳层连接,并且为了降低池化带来的梯度负面效果,作者直接摒弃了POOLing,用conv的stride来实现降采样。在这个网络结构中,使用的是步长为2的卷积来进行降采样。为了加强算法对小目标检测的精确度,YOLO v3中采用类似FPN的upsample和融合做法(最后融合了3个scale,其他两个scale的大小分别是26×26和52×52),在多个scale的feature map上做检测。作者在3条预测支路采用的也是全卷积的结构,其中最后一个卷积层的卷积核个数是255,是针对COCO数据集的80类:3*(80+4+1)=255,3表示一个grid cell包含3个bounding box,4表示框的4个坐标信息,1表示objectness score。output所谓的多尺度就是来自这3条预测之路,y1,y2和y3的深度都是255,边长的规律是13:26:52。yolo v3设定的是每个网格单元预测3个box,所以每个box需要有(x, y, w, h, confidence)五个基本参数,然后还要有80个类别的概率。所以3×(5 + 80) = 255。这个255就是这么来的。下面我们具体看看y1,y2,y3是如何而来的。网络中作者进行了三次检测,分别是在32倍降采样,16倍降采样,8倍降采样时进行检测,这样在多尺度的feature map上检测跟SSD有点像。在网络中使用up-sample(上采样)的原因:网络越深的特征表达效果越好,比如在进行16倍降采样检测,如果直接使用第四次下采样的特征来检测,这样就使用了浅层特征,这样效果一般并不好。如果想使用32倍降采样后的特征,但深层特征的大小太小,因此yolo_v3使用了步长为2的up-sample(上采样),把32倍降采样得到的feature map的大小提升一倍,也就成了16倍降采样后的维度。同理8倍采样也是对16倍降采样的特征进行步长为2的上采样,这样就可以使用深层特征进行detection。作者通过上采样将深层特征提取,其维度是与将要融合的特征层维度相同的(channel不同)。如下图所示,85层将13×13×256的特征上采样得到26×26×256,再将其与61层的特征拼接起来得到26×26×768。为了得到channel255,还需要进行一系列的3×3,1×1卷积操作,这样既可以提高非线性程度增加泛化性能提高网络精度,又能减少参数提高实时性。52×52×255的特征也是类似的过程。从图中,我们可以看出y1,y2,y3的由来。Bounding BoxYOLO v3的Bounding Box由YOLOV2又做出了更好的改进。在yolo_v2和yolo_v3中,都采用了对图像中的object采用k-means聚类。 feature map中的每一个cell都会预测3个边界框(bounding box) ,每个bounding box都会预测三个东西:(1)每个框的位置(4个值,中心坐标tx和ty,,框的高度bh和宽度bw),(2)一个objectness prediction ,(3)N个类别,coco数据集80类,voc20类。三次检测,每次对应的感受野不同,32倍降采样的感受野最大,适合检测大的目标,所以在输入为416×416时,每个cell的三个anchor box为(116 ,90); (156 ,198); (373 ,326)。16倍适合一般大小的物体,anchor box为(30,61); (62,45); (59,119)。8倍的感受野最小,适合检测小目标,因此anchor box为(10,13); (16,30); (33,23)。所以当输入为416×416时,实际总共有(52×52+26×26+13×13)×3=10647个proposal box。感受一下9种先验框的尺寸,下图中蓝色框为聚类得到的先验框。黄色框式ground truth,红框是对象中心点所在的网格。这里注意bounding box 与anchor box的区别:Bounding box它输出的是框的位置(中心坐标与宽高),confidence以及N个类别。anchor box只是一个尺度即只有宽高。LOSS FunctionYOLOv3重要改变之一:No more softmaxing the classes。YOLO v3现在对图像中检测到的对象执行多标签分类。早期YOLO,作者曾用softmax获取类别得分并用最大得分的标签来表示包含再边界框内的目标,在YOLOv3中,这种做法被修正。softmax来分类依赖于这样一个前提,即分类是相互独立的,换句话说,如果一个目标属于一种类别,那么它就不能属于另一种。但是,当我们的数据集中存在人或女人的标签时,上面所提到的前提就是去了意义。这就是作者为什么不用softmax,而用logistic regression来预测每个类别得分并使用一个阈值来对目标进行多标签预测。比阈值高的类别就是这个边界框真正的类别。用简单一点的语言来说,其实就是对每种类别使用二分类的logistic回归,即你要么是这种类别要么就不是,然后便利所有类别,得到所有类别的得分,然后选取大于阈值的类别就好了。logistic回归用于对anchor包围的部分进行一个目标性评分(objectness score),即这块位置是目标的可能性有多大。这一步是在predict之前进行的,可以去掉不必要anchor,可以减少计算量。如果模板框不是最佳的即使它超过我们设定的阈值,我们还是不会对它进行predict。不同于faster R-CNN的是,yolo_v3只会对1个prior进行操作,也就是那个最佳prior。而logistic回归就是用来从9个anchor priors中找到objectness score(目标存在可能性得分)最高的那一个。logistic回归就是用曲线对prior相对于 objectness score映射关系的线性建模。以上是一段keras框架描述的yolo v3 的loss_function代码。忽略恒定系数不看,可以从上述代码看出:除了w, h的损失函数依然采用总方误差之外,其他部分的损失函数用的是二值交叉熵。最后加到一起。那么这个binary_crossentropy又是个什么玩意儿呢?就是一个最简单的交叉熵而已,一般用于二分类,这里的两种二分类类别可以理解为"对和不对"这两种。参考文章:https://towardsdatascience.com/yolo-v3-object-detection-53fb7d3bfe6bhttps://blog.csdn.net/yanzi6969/article/details/80505421https://blog.csdn.net/chandanyan8568/article/details/81089083https://blog.csdn.net/leviopku/article/details/82660381https://blog.csdn.net/u014380165/article/details/80202337发布于 2019-03-31 10:24人工智能深度学习(Deep Learning)目标检测赞同 445 条评论分享喜欢收藏申请
DBL是什么意思? - DBL的全称 | 在线英文缩略词查询
DBL是什么意思? - DBL的全称 | 在线英文缩略词查询
↓ 跳到主内容
EnglishالعربيةБългарскиCatalàČeštinaCymraegDanskDeutschΕλληνικάEspañolEestiفارسیSuomiFrançaisעִבְרִיתहिन्दीJezikAyititMagyarBahasa IndonesiaItaliano日本語한국어LietuviųLatviešuMelayuMaltiNorskNederlandsPolskiPortuguêsRomânăРусскийSlovenčinaslovenščinaSvenskaไทยTürkçeукраїнськаاردوViệt Nam繁體中文
首页 › 3 个字母 › DBL
DBL 是什么意思?
你在寻找DBL的含义吗?在下图中,您可以看到DBL的主要定义。 如果需要,您还可以下载要打印的图像文件,或者您可以通过Facebook,Twitter,Pinterest,Google等与您的朋友分享。要查看DBL的所有含义,请向下滚动。 完整的定义列表按字母顺序显示在下表中。
DBL的主要含义
下图显示了DBL最常用的含义。 您可以将图像文件下载为PNG格式以供离线使用,或通过电子邮件发送给您的朋友。如果您是非商业网站的网站管理员,请随时在您的网站上发布DBL定义的图像。
DBL的所有定义
如上所述,您将在下表中看到DBL的所有含义。 请注意,所有定义都按字母顺序列出。您可以单击右侧的链接以查看每个定义的详细信息,包括英语和您当地语言的定义。
首字母缩写词定义DBL三角洲篮球联赛DBL下来由法律DBL不要迟到DBL代顿商界领袖DBL分布 Bernard LlechaDBL分裂的位线DBL双DBL双齿轮减速机DBL唐志强 LiluahDBL域块列表DBL基于分布的物流DBL基于磁盘的查找DBL崇 LambertDBL开发基线DBL德肖恩 B.LeroyDBL数字校样DBL数据库语言DBL数据库锁DBL杜邀请里昂DBL残疾DBL残疾福利法DBL设计兄弟有限公司DBL设计基于学习DBL达菲绑定类似DBL运球伯特兰瑟DBL透镜之间的距离DBL钻石棒球联赛
‹ DCH
DCL ›
语言
EnglishالعربيةБългарскиCatalàČeštinaCymraegDanskDeutschΕλληνικάEspañolEestiفارسیSuomiFrançaisעִבְרִיתहिन्दीJezikAyititMagyarBahasa IndonesiaItaliano日本語한국어LietuviųLatviešuMelayuMaltiNorskNederlandsPolskiPortuguêsRomânăРусскийSlovenčinaslovenščinaSvenskaไทยTürkçeукраїнськаاردوViệt Nam繁體中文
简体中文
Recent Posts
文章分类
>>
1
2
3
4
5
6
7
8
9
10
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
© 2014 - 2023
Abbreviation Finder. 站点地图 | Recent Posts
Terms of Use | Privacy Policy | About Us | Blog
Characters | Dragon Ball Legends | DBZ Space
Characters | Dragon Ball Legends | DBZ Space
NewsCharactersStorySummonsMissionsItems
close
Page
Site
Search
NewsCharactersListTagsExtrasBoosted Drop ItemsComparisonEXP LVLskeyboard_arrow_rightkeyboard_arrow_downStoryMainEventsStatus IconsPlayer Viewskeyboard_arrow_rightkeyboard_arrow_downSummonsCrystalsTicketsSuggestion Utilitykeyboard_arrow_rightkeyboard_arrow_downMissionsItemsEquipmentJob RequestsMedalsSoulsTitlesTrainingZ Powerskeyboard_arrow_rightkeyboard_arrow_downExchange ShopTraining GroundsZ LevelsHelpTournamentsResourcesGamePress LegendsOfficial Sitekeyboard_arrow_rightkeyboard_arrow_down
Settings
Display Intro Header
Reset Language
keyboard_arrow_rightkeyboard_arrow_down
ListTagsExtrasBoosted Drop ItemsComparisonEXP LVLs
Characters
- Rarity -
Ultra
Sparking
Extreme
Hero
- Color -
Red
Yellow
Purple
Green
Blue
Light
Ability Boost Mode (Beta)
Select 7-15 Characters to find the best Team Ability Boost Combinations
Super Janemba
Super Saiyan God SS Vegito
Legendary Super Saiyan Broly
Super Vegito
Ultra Instinct -Sign- Goku
Buu: Kid
Hit
Super Saiyan God SS Kaioken Goku
Super Saiyan Rosé Goku Black
Super Saiyan God SS Gogeta
Super Gogeta
Vegeta
Super Saiyan Goku
Omega Shenron
Android #17
Legends Limited
Goku & Final Form Frieza
Legends Limited
Beast Gohan
Legends Limited
Perfect Form Cell
Legends Limited
Super Saiyan Goku
Legends Limited
Gamma 1 & Gamma 2
Legends Limited
Goku
Legends Limited
Goku (Youth)
Legends Limited
Super Saiyan 3 & Super Saiyan 2 Goku & Vegeta
Legends Limited
Super Vegeta
Legends Limited
Super #17
Legends Limited
Ultra Instinct Goku
Legends Limited
Super Saiyan God SS Evolved & Super Saiyan God SS Kaioken Vegeta & Goku
Legends Limited
Super Saiyan 4 Goku & Vegeta
Legends Limited
Super Saiyan 2 & Super Saiyan God SS Trunks (Adult) & Vegeta
Legends Limited
Goku & Bardock
Legends Limited
Ultimate Gohan
Legends Limited
Android #17 & Android #18
Legends Limited
Jeice & Burter
Super Saiyan God SS Gogeta
Legends Limited
Master Roshi
Final Form Cooler
Legends Limited
Super Saiyan Trunks (Teen)
Legends Limited
Jiren: Full Power
Legends Limited
zenkaier
Super Saiyan Trunks (Adult)
Legends Limited
zenkaier
Jiren
Android #16
Super Saiyan Trunks (Adult)
Piccolo Jr.
Power Awakening Piccolo
Super Saiyan 3 Goku
Legends Limited
Perfect Cell
Legends Limited
Super Saiyan Broly: Full Power
Legends Limited
Super Saiyan 4 Gogeta
Legends Limited
zenkaier
Super Saiyan God SS Goku & Vegeta
Legends Limited
Zamasu
2nd Form Cell
Demon King Piccolo
Bardock
Legends Limited
zenkaier
Hero Tapion
Pikkon
Legends Limited
Buu: Kid
Legends Limited
Super Saiyan God SS Vegito
Legends Limited
zenkaier
Super Saiyan God SS Evolved Vegeta
Legends Limited
zenkaier
Golden Frieza
Super Saiyan Broly: Full Power
Gamma 2
Master Roshi
Super Saiyan Gohan (Youth)
Android #8
Gamma 1
Super Saiyan Gohan
Legends Limited
Mecha Frieza
Super Saiyan God SS Goku
Legends Limited
Goku
Goku
Captain Ginyu
Golden Frieza
Bulla: Vegeta (Assist)
Pan: Trunks & Giru (Assist)
God of Destruction Toppo
zenkaier
Magenta
Fusion Zamasu
Ultra Instinct -Sign- Goku
Legends Limited
zenkaier
Android #13
Super Saiyan Goku
Fused with Nail Piccolo
Kaioken Goku
zenkaier
Final Form Frieza: Full Power
Legends Limited
zenkaier
Kahseral
Super Saiyan Trunks (Teen)
Super Saiyan Rosé Goku Black
Legends Limited
zenkaier
Power Awakening Piccolo
Legends Limited
Super Saiyan Goku
Pui Pui
Super Saiyan 4 Goku
Legends Limited
zenkaier
Super Saiyan Gogeta
Broly
Android #17
Super Saiyan 2 Kefla
Super #17
Majin Vegeta
Vegeta
Legends Limited
zenkaier
Half-Corrupted Fusion Zamasu
Legends Limited
Android #17
Android #18
Android #18
Ultimate Gohan
Final Form Frieza
Final Form Frieza
Super Saiyan Broly
Tien
Saiyan in Red
Gohan
Launch
Super Saiyan Gohan (Youth)
Legends Limited
Goku
Super Saiyan 2 Kefla
Super Saiyan God SS Vegito
Super Saiyan Vegeta
Majuub
Recoome & Guldo
Fusion Zamasu
zenkaier
Gotenks
Final Form Frieza
Goku Black
Super Saiyan God SS Vegeta
Super Full Power Saiyan 4 Goku
zenkaier
Hero Tapion
Super Saiyan Trunks (Teen)
Gohan (Kid): Piccolo (Assist)
Kyawei
Nappa
Pan
1st Form Cell
Omega Shenron
zenkaier
1st Form Frieza
Jiren
zenkaier
Eis Shenron
Tien
Krillin
Final Form Cooler
Android #18
Metal Cooler
Goku Black
Ultimate Gohan Absorbed Buu: Super
Bergamo
zenkaier
Red Ribbon Army Piccolo
Super Saiyan Trunks (Adult) (Rage)
Zamasu: Goku Black (Assist)
Nuova Shenron
God of Destruction Beerus
Super Saiyan God SS Vegeta
Super Saiyan 3 Goku
Super Saiyan 3 Goku
Legends Limited
zenkaier
Perfect Form Cell
Android #21: Good
zenkaier
Trunks: Mai (Assist)
zenkaier
Whis
Boujack
Super Saiyan Gohan (Youth)
Buu: Kid
Metal Cooler
Super Saiyan 2 Caulifla: Kale (Assist)
Super Saiyan 2 Trunks (Adult)
Super Saiyan Vegeta
Super Saiyan Rosé Goku Black
Demon King Piccolo (Aged)
Super Saiyan God SS Goku
Full Power Boujack
Hit
Super Saiyan God SS Vegeta
Syn Shenron
Super Saiyan 3 Goku
God of Destruction Champa
Wasteland Bandit Yamcha
Goku
Super Saiyan God SS Goku
Super Saiyan 2 Kefla
zenkaier
Super Saiyan God SS Goku
Broly: Cheelai (Assist)
zenkaier
Legendary Super Saiyan Broly
Toppo
Android #21
zenkaier
Vegeta
Legends Limited
zenkaier
Turles
Super Saiyan God SS Vegeta
Legends Limited
zenkaier
Goku
zenkaier
Buu: Super
Goku Black
Zahha
Goku
Super Saiyan God SS Goku
Fusion Android #13
zenkaier
Super Kaioken Goku
Super Saiyan 2 Gohan (Youth)
Legends Limited
zenkaier
Super Saiyan Goku
Super Saiyan Trunks (Teen)
zenkaier
Vegito
Super Saiyan God Goku
Super Gogeta
zenkaier
Super Saiyan Rosé Goku Black
zenkaier
Super Saiyan 4 Vegeta
zenkaier
Bardock
Final Form Frieza: Full Power
zenkaier
Super Saiyan 2 Goku
Goku
zenkaier
Gotenks Absorbed Buu: Super
zenkaier
Great Saiyaman 1 & 2 (Assist)
zenkaier
Basil
Super Saiyan Goten (Kid)
Legends Limited
zenkaier
Super Janemba
zenkaier
Super Saiyan Goku
Buu: Super
Super Saiyan Gohan (Teen)
zenkaier
Lavender
Vegeta
Super Saiyan Gotenks
Bardock
Goten (Kid)
Super Baby 2
zenkaier
Super Saiyan 4 Goku
zenkaier
Golden Frieza
Vegeta
Buu: Kid
God of Destruction Beerus
Super #17
zenkaier
Ribrianne
Super Saiyan God Goku
Android #21: Evil
zenkaier
Super Saiyan Vegeta
Ultimate Gohan Absorbed Buu: Super
Gohan (Adult)
Goku
Kakarot Goku
zenkaier
Super Saiyan God SS Goku
zenkaier
Baby Vegeta
Trunks (Kid)
Raditz
1st Form Frieza
Legends Limited
zenkaier
Dyspo
Bulma (Youth)
Super Saiyan Gohan (Youth)
Android #17
zenkaier
Super Vegito
Legends Limited
zenkaier
Legendary Super Saiyan Broly
zenkaier
Android #18
Piccolo
Legends Limited
zenkaier
Ultimate Gohan Absorbed Buu: Super
1st Form Cell
Android #13
Kakunsa
Perfect Cell
zenkaier
Hyper Meta-Rilldo
Super Saiyan 3 Gotenks
zenkaier
Hero Tapion
Vados
Super Saiyan God SS Vegeta
zenkaier
Vegito
zenkaier
Super Saiyan Kale (Berserk)
zenkaier
Goku Black
Goku
Legends Limited
zenkaier
Turles
zenkaier
Hit
Final Form Frieza
Super Saiyan Gohan (Teen)
Pikkon
Gohan
zenkaier
Gogeta
zenkaier
Android #21
zenkaier
Super Saiyan 2 Caulifla
zenkaier
Great Saiyaman 1
Great Saiyaman 2
Super Saiyan Vegeta
Super Gogeta
zenkaier
Super Saiyan Vegeta
zenkaier
1st Form Frieza
Super Saiyan Goku
zenkaier
Super Saiyan Cabba
zenkaier
Ultimate Gohan
zenkaier
Super Saiyan 3 Goku
Rozie
Final Form Cooler
zenkaier
Android #14
Super Namekian Lord Slug
zenkaier
Majin Vegeta
zenkaier
Super Saiyan Goku
Metal Cooler
zenkaier
Cooler
Boujack
Super Saiyan God SS Goku
zenkaier
Super Saiyan Broly
zenkaier
Nappa (Young)
Trunks (Teen)
Majin Buu: Good
zenkaier
Super Saiyan Gohan
Android #15
Vegeta
zenkaier
Goku
Goku Captain Ginyu
Zamasu
Super Saiyan 3 Goku
zenkaier
Vegeta
Super Saiyan Gotenks
Cooler
Super Saiyan Bardock
zenkaier
Super Janemba
zenkaier
Piccolo
zenkaier
Fused with Kami Piccolo
zenkaier
Pan
zenkaier
Super Saiyan Goku
Super Saiyan 2 Trunks (Adult)
zenkaier
Final Form Frieza
zenkaier
Super Saiyan 2 Goku
zenkaier
God of Destruction Beerus
Legends Limited
zenkaier
Krillin
Super Saiyan 2 Vegeta
Super Saiyan God Goku
Vegeta
Super Saiyan Goku
Goku Black
zenkaier
Videl
Broly: Fury
zenkaier
Pikkon
Super Saiyan God Vegeta
zenkaier
King of the Demon Realm Dabura
Bunny Girl Bulma (Youth)
Captain Ginyu
Chi-Chi
zenkaier
Gotenks
zenkaier
Android #18
Super Saiyan Goku
zenkaier
Perfect Form Cell
zenkaier
Demon King Piccolo
zenkaier
Bardock
Goku (Youth)
zenkaier
Goku
zenkaier
Chilled
zenkaier
Super Saiyan 2 Gohan (Youth)
zenkaier
Super Saiyan Trunks (Kid)
Bulma (Youth)
zenkaier
Mai
Fused with Kami Piccolo
zenkaier
Android #16
zenkaier
Super Vegeta
zenkaier
Super Saiyan Goten (Kid)
Perfect Form Cell
Taitans Yamcha
Super Trunks
zenkaier
Android #18
zenkaier
Super Saiyan Goku
Legendary Super Saiyan Broly
zenkaier
Gohan (Kid)
Super Saiyan Goku
zenkaier
Trunks (Youth)
zenkaier
Goku
zenkaier
Super Saiyan Trunks (Teen)
zenkaier
World Champion Hercule
Android #18
Gohan (Kid) & Krillin
Majuub
Jiren
Omega Shenron
Super Saiyan God SS Vegeta
Thouser
Trunks
Metal Cooler
Bardock
Super Saiyan 2 Gohan (Youth)
Super Saiyan God SS Goku
Cell Jr.
God of Destruction Champa
Final Form Frieza
Gohan & Trunks (Youth)
Final Form Frieza
Vados
Fasha
Tora
Super Saiyan Kale (Berserk)
Gokua
Final Form Cooler
Bergamo
Monrei
Yurin
Turles
Legendary Super Saiyan Broly
Raspberry
Toppo
Kale
Kaiwareman
Gogeta
Kyukonman
Dyspo
Mai
Guprei
Super Saiyan 3 Gotenks
Yajirobe
Navel
Ramon
Bido
Broly
Super Saiyan 2 Goku
Nabana
Roberry
Full Power Boujack
Super Baby 2
Saibaman
Whis
Super Trunks
Appule
Super #17
Jinkouman
Nail
Zangya
Videl
Super Saiyan Goku
Majin Buu: Good
Bujin
Ribrianne
Tennenman
Buu: Super
Gohan (Kid)
Tien
Chi-Chi
Copyman
Master Roshi
Final Form Frieza
King of the Demon Realm Dabura
Orlen
Hit
Chilled
Yamcha
Baby Vegeta
Android #13
Super Saiyan Goku
Gohan (Kid)
Goku
Trunks (Youth)
Shugesh
Super Vegeta
Borgos
Recoome
Goku Black
Zangya
Majin Buu: Pure Evil
zenkaier
Captain Ginyu
zenkaier
Krillin
Super Saiyan Rosé Goku Black
Kakunsa
Zamasu
Android #14
2nd Form Frieza
Hyper Meta-Rilldo
Goku (Youth)
Videl
Krillin
zenkaier
Syn Shenron
Android #15
Trunks
Boujack
Rozie
Trunks (Kid)
3rd Form Frieza
Super Janemba
Perfect Form Cell
Super Namekian Lord Slug
Jeice
Shisami
God of Destruction Beerus
zenkaier
1st Form Frieza
zenkaier
Trunks (Teen)
Vegeta
zenkaier
Bulma (Youth)
Tien
Demon King Piccolo
Gohan
zenkaier
Super Saiyan Vegeta
Burter
Tora
Turles
Android #17
Android #18
Tien
Vegeta
Pan: Honey
Android #16
Guldo
Caulifla
Fasha
2nd Form Cell
zenkaier
Fused with Nail Piccolo
Krillin
Super Saiyan Trunks (Teen)
Mai
Cell Jr.
Gohan (Youth)
Super Saiyan Vegeta
Yamcha
Cooler
Thouser
Cabba
Trunks (Kid)
Mercenary Tao
Goten (Kid)
Dodoria
Final Form Frost
Master Roshi
Bardock
1st Form Frieza
Chiaotzu
Zarbon
Vegeta
Pan
Android #18
zenkaier
Android #17
zenkaier
Raditz
zenkaier
Goku
Android #20
Nappa
zenkaier
Android #19
Trunks (Teen)
1st Form Cell
zenkaier
Videl
Mercenary Tao
Nail
Thouser
Turles
Krillin
Super Saiyan Goku
2nd Form Cell
World Champion Hercule
Nappa
1st Form Cell
Goku
Fused with Nail Piccolo
Raditz
Recoome
Shallot
Vegeta
Pan
Krillin
Cabba
Captain Ginyu
Cell Jr.
Tien
Zarbon
Jaco
Goku
2nd Form Frieza
Chiaotzu
Master Roshi
Final Form Frost
Saibaman
Jaco
Guldo
Saibaman
Gohan (Kid)
1st Form Frieza
Yamcha
Yamcha
Dodoria
Android #19
Jeice
Burter
Android #20
Pikkon
Load More
0 / 15 Characters
Reset
Next
Privacy Policy
DRAGON BALL LEGENDS|Bandai Namco Entertainment Official Site
DRAGON BALL LEGENDS|Bandai Namco Entertainment Official Site
CHARACTERS
NEWS
MEDIA
TUTORIALS
STORY
Bandai Namco ID
EN
JP
EN
FR
DE
ES
KR
TW
DOWNLOAD FOR FREE*
60M+
PLAYERS WORLDWIDE
1M+
SOCIAL MEDIA FOLLOWERS
400+
AMAZING 3D CHARACTERS
News
100+ CHARACTERS
Collect hundreds of characters from different sagas
PvP BATTLES
Compete in real-time battles against fans from around the world.
Discover Our Game Modes
Story ModeLearn more
Co-op BattlesLearn more
Tournament of POWERLearn more
Tutorial Series
TAKE THE TEST
From beginnerto expert!
FAQ CORNER
Get your questionsanswered NOW!
VIDEO TIPS
Watch videolessons NOW!
Latest News
To view this content, please allow our site to use cookies.Click here to adjust your Cookie Settings.
DOWNLOAD NOW!
DRAGON BALL LEGENDS
Home
CHARACTERS
NEWS
MEDIA
TUTORIALS
STORY
Bandai Namco ID
LEGAL
iOS Support
Android Support
Terms of Service
Privacy Policy
DRAGON BALL LEGENDS
SHARE WITH FRIENDS
In-app purchases available.
Apple and the Apple logo are trademarks of Apple Inc., registered in the U.S. and other countries and regions.
App Store is a service mark of Apple Inc.
Google Play and the Google Play logo are trademarks of Google LLC.
©BIRD STUDIO/SHUEISHA, TOEI ANIMATION©Bandai Namco Entertainment Inc.
Dragon Ball Legends | DBZ Space
Dragon Ball Legends | DBZ Space
NewsCharactersStorySummonsMissionsItems
close
Page
Site
Search
NewsCharactersListTagsExtrasBoosted Drop ItemsComparisonEXP LVLskeyboard_arrow_rightkeyboard_arrow_downStoryMainEventsStatus IconsPlayer Viewskeyboard_arrow_rightkeyboard_arrow_downSummonsCrystalsTicketsSuggestion Utilitykeyboard_arrow_rightkeyboard_arrow_downMissionsItemsEquipmentJob RequestsMedalsSoulsTitlesTrainingZ Powerskeyboard_arrow_rightkeyboard_arrow_downExchange ShopTraining GroundsZ LevelsHelpTournamentsResourcesGamePress LegendsOfficial Sitekeyboard_arrow_rightkeyboard_arrow_down
Settings
Display Intro Header
Reset Language
keyboard_arrow_rightkeyboard_arrow_down
News
Event
"Treasure Battle" Begins!
02/20/2024 11:30 PM (CST)
~
12/31/2029 09:00 AM (CST)
03/05/2024 11:35 PM (CST)
Important
Important: Data Transfer Settings
Click here if you want to change your device
02/20/2024 11:30 PM (CST)
~
12/31/2029 09:00 AM (CST)
02/21/2024 02:55 AM (CST)
News
About Equipment
02/20/2024 11:30 PM (CST)
~
12/31/2029 09:00 AM (CST)
02/21/2024 02:55 AM (CST)
Event
"Vegeta, Prince of All Saiyans - Rebellion Against Frieza! -" On Now!
10/20/2021 12:30 AM (CDT)
~
12/31/2029 09:00 AM (CST)
11/28/2023 11:40 PM (CST)
Event
"Eliminate the Earthlings" Now On!
11/18/2020 12:00 AM (CST)
~
12/31/2029 09:00 AM (CST)
11/28/2023 11:40 PM (CST)
Event
Play "Legends Road - Gotenks -" Whenever You Want!
12/22/2021 12:00 AM (CST)
~
12/31/2029 09:00 AM (CST)
11/28/2023 11:40 PM (CST)
Summon
"LET'S GO LEGENDS!! LEGENDS LIMITED Guaranteed Ticket Summon" Is Now On!
05/18/2022 01:00 AM (CDT)
~
12/31/2029 09:00 AM (CST)
11/28/2023 11:40 PM (CST)
Event
From the Original Series "The Dreaded Ginyu Force" Is Now Available In SCREAM Difficulty!
07/21/2021 01:00 AM (CDT)
~
12/31/2029 09:00 AM (CST)
11/21/2023 11:35 PM (CST)
Event
From the Original Series "The Advent of Frieza, the Universal Emperor!" Overhaul!
12/09/2020 12:00 AM (CST)
~
12/31/2029 09:00 AM (CST)
11/21/2023 05:30 AM (CST)
Event
From the Original Series "Fierce Battle! Vegeta the Prince of Fate" Overhaul!
11/11/2020 12:00 AM (CST)
~
12/31/2029 09:00 AM (CST)
11/21/2023 05:30 AM (CST)
Older
Privacy Policy
Dragon Ball Legends Database
Dragon Ball Legends Database
Main Menu
Home
News
Characters
Summons
Events
Items
Equipment
Messages
Missions
Extras
Help
Links
Hydros' Twitter
Join the Discord
Subreddit
Dragon Ball Legends Wiki France
Widen Legends ES
MyHeroUI Database
MyHeroUR Database
Dokkan Wiki
Official Global Twitter
Official Japan Twitter
Language
English
日本語
Français
Español
Deutsch
中文
한국어
Latest Characters
Broly: Fury
Golden Frieza
Super Saiyan 2 Gohan (Youth)
Perfect Form Cell
Perfect Cell
LEGENDS LIMITED
Super Vegeta
Android #16
Latest Equipment
Latest Summons
Latest Events
Latest News
DBLegends.net v0.8.4a
Privacy Policy
Patreon