📄 simple object detector with boosting.mht
字号:
From: <由 Windows Internet Explorer 7 保存>
Subject: Simple object detector with boosting
Date: Sun, 3 Feb 2008 20:10:25 +0800
MIME-Version: 1.0
Content-Type: multipart/related;
type="text/html";
boundary="----=_NextPart_000_0000_01C866A0.D0018460"
X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2900.3198
This is a multi-part message in MIME format.
------=_NextPart_000_0000_01C866A0.D0018460
Content-Type: text/html;
charset="utf-8"
Content-Transfer-Encoding: quoted-printable
Content-Location: http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/boosting.html
=EF=BB=BF<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD><TITLE>Simple object detector with boosting</TITLE>
<META http-equiv=3DContent-Type content=3D"text/html; charset=3Dutf-8">
<META content=3D"MSHTML 6.00.6000.16587" name=3DGENERATOR></HEAD>
<BODY text=3D#333300 vLink=3D#003300 aLink=3D#003300 link=3D#006600 =
bgColor=3D#ffffff>
<TABLE style=3D"WIDTH: 800px" width=3D800=20
background=3Dhttp://people.csail.mit.edu/torralba/shortCourseRLOC/boostin=
g/=20
border=3D0>
<TBODY>
<TR>
<TD width=3D200>
<DIV align=3Dleft><IMG height=3D200=20
=
src=3D"http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/boos=
tingIcon.gif"=20
width=3D200 border=3D0> </DIV><BR></TD>
<TD width=3D20></TD>
<TD width=3D780><BR><FONT size=3D6><B>A simple object detector =
<BR>with=20
boosting</B><BR></FONT><BR><FONT size=3D4><B>ICCV 2005 short =
courses on=20
<BR><A=20
=
href=3D"http://people.csail.mit.edu/torralba/iccv2005/index.html">Recogni=
zing=20
and Learning Object Categories=20
</A></B><BR></FONT><BR><BR></TD></TR></TBODY></TABLE><BR>
<TABLE style=3D"WIDTH: 800px" width=3D800=20
background=3Dhttp://people.csail.mit.edu/torralba/shortCourseRLOC/boostin=
g/=20
border=3D0>
<TBODY>
<TR>
<TD align=3Djustify width=3D1000><FONT size=3D3>Boosting provides a =
simple=20
framework to develop robust object detection algorithms. This set =
of=20
functions provide a minimal set to build an object detection =
algorithm. It=20
is entirely written on Matlab in order to make it easily accesible =
as a=20
teaching tool. Therefore, it is not appropriate for building =
real-time=20
applications. </FONT><BR></TD></TR></TBODY></TABLE><BR>
<TABLE style=3D"WIDTH: 800px" width=3D800=20
background=3Dhttp://people.csail.mit.edu/torralba/shortCourseRLOC/boostin=
g/=20
border=3D0>
<TBODY>
<TR>
<TD width=3D1000>
<HR align=3Dleft width=3D800 SIZE=3D1>
<FONT size=3D6>Setup </FONT><BR><BR><A=20
=
href=3D"http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/boo=
stingDemo.zip">Download</A>=20
the code and datasets<BR><A=20
=
href=3D"http://people.csail.mit.edu/torralba/LabelMeToolbox/LabelMeToolbo=
x.zip">Download</A>=20
the LabelMe toolbox<BR><BR>Unzip both files. Modify the paths in =
<FONT=20
color=3D#008800>initpath.m</FONT><BR>Modify the folder paths in =
<FONT=20
color=3D#008800>paramaters.m</FONT> to point to the locations of =
the images=20
and annotations.<BR><BR>
<HR align=3Dleft width=3D800 SIZE=3D1>
<FONT size=3D6>Description of the functions </FONT><BR><BR><FONT=20
size=3D3><B>Initialization</B><BR><FONT =
color=3D#008800>initpath.m</FONT> -=20
Initializes the matlab path. You should run this command when you =
start=20
the Matlab session.<BR><FONT color=3D#008800>paremeters.m</FONT> - =
Contains=20
parameters to configure the classifiers and the database.<BR>
<HR align=3Dleft width=3D800 SIZE=3D1>
<FONT size=3D3><B>Boosting tools</B><BR><FONT=20
color=3D#008800>demoGentleBoost.m</FONT> - simple demo of =
gentleBoost using=20
stumps on two dimensions<BR>
<HR align=3Dleft width=3D800 SIZE=3D1>
<FONT size=3D3><B>Scripts</B><BR><FONT=20
color=3D#008800>createDatabases.m</FONT> - creates the training =
and test=20
database using the LabelMe database.<BR><FONT=20
color=3D#008800>createDictionary.m</FONT> - creates a dictionary =
of filtered=20
patches from the target object.<BR><FONT=20
color=3D#008800>computeFeatures.m</FONT> - precomputes the =
features of all=20
images and stores the feature outputs on the center of the target =
object=20
and on a sparse set of locations from the background.<BR><FONT=20
color=3D#008800>trainDetector.m</FONT> - creates the training and =
test=20
database using the LabelMe database<BR><FONT=20
color=3D#008800>runDetector.m</FONT> - runs the detector on test =
images<BR>
<HR align=3Dleft width=3D800 SIZE=3D1>
<B>Features and weak detectors</B><BR><FONT=20
color=3D#008800>convCrossConv.m</FONT> - Weak detector: computes =
template=20
matching with a localized patch in object centered =
coordinates.<BR>
<HR align=3Dleft width=3D800 SIZE=3D1>
<FONT size=3D3><B>Detector</B><BR><FONT=20
color=3D#008800>singleScaleBoostedDetector.m</FONT> - runs the =
strong=20
classifier on an image at a single scale and outputs bounding =
boxes and=20
scores.<BR>
<HR align=3Dleft width=3D800 SIZE=3D1>
<B>LabelMe toolbox </B><BR><A=20
=
href=3D"http://people.csail.mit.edu/torralba/LabelMeToolbox/">LabelMe</A>=
-=20
Describes the utility functions used to manipulate the=20
database<BR><BR></FONT>
<DIV></DIV></FONT></FONT></FONT></TD></TR></TBODY></TABLE>
<TABLE style=3D"WIDTH: 800px" width=3D800=20
background=3Dhttp://people.csail.mit.edu/torralba/shortCourseRLOC/boostin=
g/=20
border=3D0>
<TBODY>
<TR>
<TD width=3D1000>
<HR align=3Dleft width=3D800 SIZE=3D1>
<FONT size=3D6>Examples </FONT><BR><BR><B>Setup</B><BR>First run =
<FONT=20
color=3D#008800>initpath.m</FONT> and modify the folder paths in =
the script=20
<FONT color=3D#008800>parameters.m</FONT> =
<BR><BR><B>Boosting</B><BR>First=20
run the Boosting demo <FONT=20
color=3D#008800>demoGentleBoost.m</FONT><BR><BR>This demo will =
first ask for=20
a set of points in 2D to be used a training data (Left button =3D =
class +1,=20
right button =3D class -1). The classifier will only be able to =
perform=20
simple discrimination tasks as it uses stumps as weak classifiers =
(i.e.,=20
only lines parallel to the axis). If you use weak classifiers to =
be lines=20
with any orientation, then you will get more interesting =
boundaries=20
easily. However, stumps are frequently used in object detection as =
they=20
can be used to do efficient feature selection. This demo will show =
you the=20
limits of stumps. In object detection, some of these limitations =
are=20
compensated by using a very large number of features.<BR><IMG =
height=3D420=20
=
src=3D"http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/demo=
Boosting.jpg"=20
width=3D560 border=3D0> <BR><B>A look to the database</B><BR>This =
is a sample=20
of the images used for this demo. They contain cars (side views) =
and=20
screens (frontal views), with normalized scale. They are a small =
subset of=20
the LabelMe dataset. The program <FONT=20
color=3D#008800>createDatabase.m</FONT> shows how the database =
used for this=20
demo was created. <BR><IMG height=3D575=20
=
src=3D"http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/exam=
plesDatabase.jpg"=20
width=3D716 border=3D0> <BR>If you download the full database, the =
first thing=20
you have to do is to actualize the folders in <FONT=20
color=3D#008800>parameters.m</FONT>. Then, you have to run the =
program <FONT=20
color=3D#008800>createDatabase.m</FONT> which will read all the =
annotation=20
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -