Making sense of LayoutInflater

It’s common practice in the Android SDK, but you may be surprised to find that there is a wrong way to use LayoutInflater, and your application might be one of the offenders. If you’ve ever written something like the following code usingLayoutInflater in your Android application:

PLEASE read on, because you’re doing it wrong and I want to explain to you why.

Get to Know LayoutInflater

Let’s first take a look at how LayoutInflater works. There are two usable versions of the inflate() method for a standard application:

The first parameter points to the layout resource you want to inflate. The second parameter is the root view of the hierarchy you are inflating the resource to attach to. When the third parameter is present, it governs whether or not the inflated view is attached to the supplied root after inflation.

It is these last two parameters that can cause a bit of confusion. With the two parameter version of this method,LayoutInflater will automatically attempt to attach the inflated view to the supplied root. However, the framework has a check in place that if you pass null for the root it bypasses this attempt to avoid an application crash.

Many developers take this behavior to mean that the proper way to disable attachment on inflation is by passing null as root; in many cases not even realizing that the three parameter version of inflate() exists. By doing things this way, we also disable another very important function the root view has…but I’m getting ahead of myself.

Examples from the Framework

Let’s examine some situations in Android where the framework expects you as a developer to interactively inflate portions of the view.

Adapters are the most common case for using LayoutInflater is custom ListView adapters overriding getView(), which has the following method signature:

Fragments also use inflation often when creating views via onCreateView(); notice its method signature:

Have you noticed that every time the framework wants you to inflate a layout, they also pass you the parent ViewGroup it will eventually be attached to? Notice also that in most cases (including the above two examples), it will throw an Exception later on if LayoutInflater is allowed to automatically attach the inflated view to the root.

So why do you suppose we are given this ViewGroup if we are not supposed to attach to it? It turns out the parent view is a very important part of the inflation process because it is necessary in order to evaluate the LayoutParams declared in the root element of the XML being inflated. Passing nothing here is akin to telling the framework “I don’t know what parent this view will be attached to, sorry.”

The problem with this is android:layout_xxx attributes are always be evaluated in the context of the parent view. As a result, without any known parent, all LayoutParams you declared on the root element of your XML tree will just get thrown away, and then you’ll be left asking “why is the framework ignoring the layout customizations I defined? I’d better check SO and then file a bug.”

Without LayoutParams, the ViewGroup that eventually hosts the inflated layout is left to generate a default set for you. If you are lucky (and in many cases you are) these default parameters are the same as what you had in XML…masking the fact that something is amiss.

Application Example

So you claim you’ve never seen this happen in an application? Take the following simple layout that we want to inflate for aListView row:

R.layout.item_row

We want to set the height of our row to be a fixed height, in this case the preferred item height for the current theme…seems reasonable.

However, when we inflate this layout the wrong way

we end up with a result that looks like this

What happened to the fixed height we set?? This is usually where you end up setting the fixed height on all your child views, switching the root elements height to wrap_content, and move on without really knowing why it broke (you may have even cursed at Google in the process).

If we instead inflate the same layout this way

we end up with what we expected in the first place.

Hooray!

Every Rule Has An Exception

There are of course instances where you can truly justify a null parent during inflation, but they are few. One such instance occurs when you are inflating a custom layout to be attached to an AlertDialog. Consider the following example where we want to use our same XML layout but set it as the dialog view:

The issue here is that AlertDialog.Builder supports a custom view, but does not provide an implementation of setView() that takes a layout resource; so you must inflate the XML manually. However, because the result will go into the dialog, which does not expose its root view (in fact, it doesn’t exist yet), we do not have access to the eventual parent of the layout, so we cannot use it for inflation. It turns out, this is irrelevant, because AlertDialog will erase any LayoutParams on the layout anyway and replace them with match_parent.

I've investigated this issue, referring to the LayoutInflater docs and setting up a small sample demonstration project. The following tutorials shows how to dynamically populate a layout using LayoutInflater.

Before we get started see what LayoutInflater.inflate() parameters look like:

  • resource: ID for an XML layout resource to load (e.g., R.layout.main_page)
  • root: Optional view to be the parent of the generated hierarchy (if attachToRoot is true), or else simply an object that provides a set of LayoutParams values for root of the returned hierarchy (if attachToRoot is false.)
  • attachToRoot: Whether the inflated hierarchy should be attached to the root parameter? If false, root is only used to create the correct subclass of LayoutParams for the root view in the XML.

  • Returns: The root View of the inflated hierarchy. If root was supplied and attachToRoot is true, this is root; otherwise it is the root of the inflated XML file.

Now for the sample layout and code.

Main layout (main.xml):

<?xml version="1.0" encoding="utf-8"?>
<LinearLayout xmlns:android="http://schemas.android.com/apk/res/android"
    android:id="@+id/container"
    android:layout_width="match_parent"
    android:layout_height="match_parent">
</LinearLayout>

Added into this container is a separate TextView, visible as small red square if layout parameters are successfully applied from XML (red.xml):

<?xml version="1.0" encoding="utf-8"?>
<TextView xmlns:android="http://schemas.android.com/apk/res/android"
    android:layout_width="25dp"
    android:layout_height="25dp"
    android:background="#ff0000"
    android:text="red" />

Now LayoutInflater is used with several variations of call parameters

public class InflaterTest extends Activity {

    private View view;

    @Override
    public void onCreate(Bundle savedInstanceState) {
      super.onCreate(savedInstanceState);

      setContentView(R.layout.main);
      ViewGroup parent = (ViewGroup) findViewById(R.id.container);

      // result: layout_height=wrap_content layout_width=match_parent
      view = LayoutInflater.from(this).inflate(R.layout.red, null);
      parent.addView(view);

      // result: layout_height=100 layout_width=100
      view = LayoutInflater.from(this).inflate(R.layout.red, null);
      parent.addView(view, 100, 100);

      // result: layout_height=25dp layout_width=25dp
      // view=textView due to attachRoot=false
      view = LayoutInflater.from(this).inflate(R.layout.red, parent, false);
      parent.addView(view);

      // result: layout_height=25dp layout_width=25dp 
      // parent.addView not necessary as this is already done by attachRoot=true
      // view=root due to parent supplied as hierarchy root and attachRoot=true
      view = LayoutInflater.from(this).inflate(R.layout.red, parent, true);
    }
}

The actual results of the parameter variations are documented in the code.

SYNOPSIS: Calling LayoutInflater without specifying root leads to inflate call ignoring the layout parameters from the XML. Calling inflate with root not equal null and attachRoot=true does load the layout parameters, but returns the root object again, which prevents further layout changes to the loaded object (unless you can find it using findViewById()). The calling convention you most likely would like to use is therefore this one:

loadedView = LayoutInflater.from(context)
                .inflate(R.layout.layout_to_load, parent, false);

  • 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