Overview of COCOMO

The COCOMO cost estimation model is used by thousands of software project managers, and is based on a study of hundreds of software projects. Unlike other cost estimation models, COCOMO is an open model, so all of the details are published, including:

  • The underlying cost estimation equations
  • Every assumption made in the model (e.g. "the project will enjoy good management")
  • Every definition (e.g. the precise definition of the Product Design phase of a project)
  • The costs included in an estimate are explicitly stated (e.g. project managers are included, secretaries aren't)

Because COCOMO is well defined, and because it doesn't rely upon proprietary estimation algorithms, Costar offers these advantages to its users:

  • COCOMO estimates are more objective and repeatable than estimates made by methods relying on proprietary models
  • COCOMO can be calibrated to reflect your software development environment, and to produce more accurate estimates

Costar is a faithful implementation of the COCOMO model that is easy to use on small projects, and yet powerful enough to plan and control large projects.

Typically, you'll start with only a rough description of the software system that you'll be developing, and you'll use Costar to give you early estimates about the proper schedule and staffing levels. As you refine your knowledge of the problem, and as you design more of the system, you can use Costar to produce more and more refined estimates.

Costar allows you to define a software structure to meet your needs. Your initial estimate might be made on the basis of a system containing 3,000 lines of code. Your second estimate might be more refined so that you now understand that your system will consist of two subsystems (and you'll have a more accurate idea about how many lines of code will be in each of the subsystems). Your next estimate will continue the process -- you can use Costar to define the components of each subsystem. Costar permits you to continue this process until you arrive at the level of detail that suits your needs.

One word of warning: It is so easy to use Costar to make software cost estimates, that it's possible to misuse it -- every Costar user should spend the time to learn the underlying COCOMO assumptions and definitions from Software Engineering Economics and Software Cost Estimation with COCOMO II.

Introduction to the COCOMO Model

The most fundamental calculation in the COCOMO model is the use of the Effort Equation to estimate the number of Person-Months required to develop a project. Most of the other COCOMO results, including the estimates for Requirements and Maintenance, are derived from this quantity.

Source Lines of Code

The COCOMO calculations are based on your estimates of a project's size in Source Lines of Code (SLOC). SLOC is defined such that:

  • Only Source lines that are DELIVERED as part of the product are included -- test drivers and other support software is excluded
  • SOURCE lines are created by the project staff -- code created by applications generators is excluded
  • One SLOC is one logical line of code
  • Declarations are counted as SLOC
  • Comments are not counted as SLOC

The original COCOMO 81 model was defined in terms of Delivered Source Instructions, which are very similar to SLOC.  The major difference between DSI and SLOC is that a single Source Line of Code may be several physical lines.  For example, an "if-then-else" statement would be counted as one SLOC, but might be counted as several DSI.

The Scale Drivers

In the COCOMO II model, some of the most important factors contributing to a project's duration and cost are the Scale Drivers. You set each Scale Driver to describe your project; these Scale Drivers determine the exponent used in the Effort Equation.

The 5 Scale Drivers are:

  • Precedentedness
  • Development Flexibility
  • Architecture / Risk Resolution
  • Team Cohesion
  • Process Maturity

Note that the Scale Drivers have replaced the Development Mode of COCOMO 81.  The first two Scale Drivers, Precedentedness and Development Flexibility actually describe much the same influences that the original Development Mode did.

Cost Drivers

COCOMO II has 17 cost drivers ?you assess your project, development environment, and team to set each cost driver. The cost drivers are multiplicative factors that determine the effort required to complete your software project. For example, if your project will develop software that controls an airplane's flight, you would set the Required Software Reliability (RELY) cost driver to Very High. That rating corresponds to an effort multiplier of 1.26, meaning that your project will require 26% more effort than a typical software project.

Click here to see which Cost Drivers are in which Costar models.

COCOMO II defines each of the cost drivers, and the Effort Multiplier associated with each rating. Check the Costar help for details about the definitions and how to set the cost drivers.

COCOMO II Effort Equation

The COCOMO II model makes its estimates of required effort (measured in Person-Months ?PM) based primarily on your estimate of the software project's size (as measured in thousands of SLOC, KSLOC)):

  
  Effort = 2.94 * EAF * (KSLOC)E

Where
    EAF   Is the Effort Adjustment Factor derived from the Cost Drivers
    E        Is an exponent derived from the five Scale Drivers

As an example, a project with all Nominal Cost Drivers and Scale Drivers would have an EAF of 1.00 and exponent, E, of 1.0997. Assuming that the project is projected to consist of 8,000 source lines of code, COCOMO II estimates that 28.9 Person-Months of effort is required to complete it:

    Effort = 2.94 * (1.0) * (8)1.0997 = 28.9 Person-Months

Effort Adjustment Factor

The Effort Adjustment Factor in the effort equation is simply the product of the effort multipliers corresponding to each of the cost drivers for your project.

For example, if your project is rated Very High for Complexity (effort multiplier of 1.34), and Low for Language & Tools Experience (effort multiplier of 1.09), and all of the other cost drivers are rated to be Nominal (effort multiplier of 1.00), the EAF is the product of 1.34 and 1.09.

    Effort Adjustment Factor = EAF = 1.34 * 1.09 = 1.46

    Effort = 2.94 * (1.46) * (8)1.0997 = 42.3 Person-Months

COCOMO II Schedule Equation

The COCOMO II schedule equation predicts the number of months required to complete your software project. The duration of a project is based on the effort predicted by the effort equation:

    Duration = 3.67 * (Effort)SE

Where
    Effort   Is the effort from the COCOMO II effort equation
    SE        Is the schedule equation exponent derived from the five Scale Drivers

Continuing the example, and substituting the exponent of 0.3179 that is calculated from the scale drivers, yields an estimate of just over a year, and an average staffing of between 3 and 4 people:

    Duration = 3.67 * (42.3)0.3179 = 12.1 months

    Average staffing = (42.3 Person-Months) / (12.1 Months) = 3.5 people

The SCED Cost Driver

The COCOMO cost driver for Required Development Schedule (SCED) is unique, and requires a special explanation.

The SCED cost driver is used to account for the observation that a project developed on an accelerated schedule will require more effort than a project developed on its optimum schedule. A SCED rating of Very Low corresponds to an Effort Multiplier of 1.43 (in the COCOMO II.2000 model) and means that you intend to finish your project in 75% of the optimum schedule (as determined by a previous COCOMO estimate). Continuing the example used earlier, but assuming that SCED has a rating of Very Low, COCOMO produces these estimates:

    Duration = 75% * 12.1 Months = 9.1 Months

    Effort Adjustment Factor = EAF = 1.34 * 1.09 * 1.43 = 2.09

    Effort = 2.94 * (2.09) * (8)1.0997 = 60.4 Person-Months


    Average staffing = (60.4 Person-Months) / (9.1 Months) = 6.7 people

Notice that the calculation of duration isn't based directly on the effort (number of Person-Months) ?instead it's based on the schedule that would have been required for the project assuming it had been developed on the nominal schedule. Remember that the SCED cost driver means "accelerated from the nominal schedule".

The Costar command Constraints | Constrain Project displays a dialog box that lets you trade off duration vs. effort (SCED is set for you automatically).  You can use the dialog box to constrain your project to have a fixed duration, or a fixed cost.

  • 0
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
1 目标检测的定义 目标检测(Object Detection)的任务是找出图像中所有感兴趣的目标(物体),确定它们的类别和位置,是计算机视觉领域的核心问题之一。由于各类物体有不同的外观、形状和姿态,加上成像时光照、遮挡等因素的干扰,目标检测一直是计算机视觉领域最具有挑战性的问题。 目标检测任务可分为两个关键的子任务,目标定位和目标分类。首先检测图像中目标的位置(目标定位),然后给出每个目标的具体类别(目标分类)。输出结果是一个边界框(称为Bounding-box,一般形式为(x1,y1,x2,y2),表示框的左上角坐标和右下角坐标),一个置信度分数(Confidence Score),表示边界框中是否包含检测对象的概率和各个类别的概率(首先得到类别概率,经过Softmax可得到类别标签)。 1.1 Two stage方法 目前主流的基于深度学习的目标检测算法主要分为两类:Two stage和One stage。Two stage方法将目标检测过程分为两个阶段。第一个阶段是 Region Proposal 生成阶段,主要用于生成潜在的目标候选框(Bounding-box proposals)。这个阶段通常使用卷积神经网络(CNN)从输入图像中提取特征,然后通过一些技巧(如选择性搜索)来生成候选框。第二个阶段是分类和位置精修阶段,将第一个阶段生成的候选框输入到另一个 CNN 中进行分类,并根据分类结果对候选框的位置进行微调。Two stage 方法的优点是准确度较高,缺点是速度相对较慢。 常见Tow stage目标检测算法有:R-CNN系列、SPPNet等。 1.2 One stage方法 One stage方法直接利用模型提取特征值,并利用这些特征值进行目标的分类和定位,不需要生成Region Proposal。这种方法的优点是速度快,因为省略了Region Proposal生成的过程。One stage方法的缺点是准确度相对较低,因为它没有对潜在的目标进行预先筛选。 常见的One stage目标检测算法有:YOLO系列、SSD系列和RetinaNet等。 2 常见名词解释 2.1 NMS(Non-Maximum Suppression) 目标检测模型一般会给出目标的多个预测边界框,对成百上千的预测边界框都进行调整肯定是不可行的,需要对这些结果先进行一个大体的挑选。NMS称为非极大值抑制,作用是从众多预测边界框中挑选出最具代表性的结果,这样可以加快算法效率,其主要流程如下: 设定一个置信度分数阈值,将置信度分数小于阈值的直接过滤掉 将剩下框的置信度分数从大到小排序,选中值最大的框 遍历其余的框,如果和当前框的重叠面积(IOU)大于设定的阈值(一般为0.7),就将框删除(超过设定阈值,认为两个框的里面的物体属于同一个类别) 从未处理的框中继续选一个置信度分数最大的,重复上述过程,直至所有框处理完毕 2.2 IoU(Intersection over Union) 定义了两个边界框的重叠度,当预测边界框和真实边界框差异很小时,或重叠度很大时,表示模型产生的预测边界框很准确。边界框A、B的IOU计算公式为: 2.3 mAP(mean Average Precision) mAP即均值平均精度,是评估目标检测模型效果的最重要指标,这个值介于0到1之间,且越大越好。mAP是AP(Average Precision)的平均值,那么首先需要了解AP的概念。想要了解AP的概念,还要首先了解目标检测中Precision和Recall的概念。 首先我们设置置信度阈值(Confidence Threshold)和IoU阈值(一般设置为0.5,也会衡量0.75以及0.9的mAP值): 当一个预测边界框被认为是True Positive(TP)时,需要同时满足下面三个条件: Confidence Score > Confidence Threshold 预测类别匹配真实值(Ground truth)的类别 预测边界框的IoU大于设定的IoU阈值 不满足条件2或条件3,则认为是False Positive(FP)。当对应同一个真值有多个预测结果时,只有最高置信度分数的预测结果被认为是True Positive,其余被认为是False Positive。 Precision和Recall的概念如下图所示: Precision表示TP与预测边界框数量的比值 Recall表示TP与真实边界框数量的比值 改变不同的置信度阈值,可以获得多组Precision和Recall,Recall放X轴,Precision放Y轴,可以画出一个Precision-Recall曲线,简称P-R
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值