simple object detector with boosting.mht

MIT的一个adaboost算法演示程序

      files and will create a struct that will be used later by the =
query tools.=20
      For more information about how the query tools work, you can check =
the <A=20
      =
href=3D"http://people.csail.mit.edu/torralba/LabelMeToolbox/">LabelMe=20
      Toolbox</A>.<BR><BR><B>Running the detector</B><BR>Before trying =
to train=20
      your own detector, you can try the script <FONT=20
      color=3D#008800>runDetector.m</FONT>. If everything is setup =
right, the=20
      output should look like: <BR><IMG height=3D147=20
      =
src=3D"http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/dete=
ctorOutput.bmp"=20
      width=3D671 border=3D0> <BR><IMG height=3D229=20
      =
src=3D"http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/dete=
ctorPrecisionRecall.bmp"=20
      width=3D671 border=3D0> <BR>Here there is an example of the output =
of the=20
      detector when trained to detected side views of cars: <BR><IMG =
height=3D198=20
      =
src=3D"http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/carD=
etectorOutput.bmp"=20
      width=3D811 border=3D0> <BR><B>Training a new detector</B><BR>To =
train a new=20
      detector, first you need to collect a new set of images. If you =
use the=20
      full LabelMe database, then, you will only need to change the =
object name=20
      in the program <FONT color=3D#008800>parameters.m</FONT> to =
indicate the=20
      object category you want to detect. Also, in <FONT=20
      color=3D#008800>parameters.m</FONT> you can change training =
parameters such=20
      as the number of training images, the size of the patches, the =
scale of=20
      the object, the number of negative examples, etc.<BR><BR><FONT=20
      color=3D#008800>createDictionary.m</FONT> will create the =
vocabulary of=20
      patches used to compute the features.<BR><BR><FONT=20
      color=3D#008800>computeFeatures.m</FONT> will precompute all the =
features=20
      for the training images.<BR><BR><FONT =
color=3D#008800>trainDetector.m</FONT>=20
      will train the detector using Gentle Boosting [1].<BR><BR>Every =
one of=20
      these programs adds information to the 'data' struct which will =
contain=20
      information such as the precomputed features, list of images used =
for=20
      training, the dictionary of features, the parameters of the =
classifier.=20
      <BR><BR>Finally, with <FONT color=3D#008800>runDetector.m</FONT> =
you can run=20
      the new detector.<BR><BR><B>Multiscale detector</B><BR><BR>In =
order to=20
      build a multiscale detectors, you need to loop on scales. =
Something like=20
      this: <FONT color=3D#008800><BR>scalingStep =3D 0.8;<BR>for scale =
=3D=20
      1:Nscales<BR>&nbsp;&nbsp;&nbsp;img =3D imresize(img, scalingStep,=20
      'bilinear');<BR>&nbsp;&nbsp;&nbsp;[Score{scale}, =
boundingBox{scale},=20
      boxScores{scale}] =3D singleScaleBoostedDetector(img,=20
      data);<BR>end<BR></FONT><BR><BR>
      <HR align=3Dleft width=3D800 SIZE=3D1>
      <FONT size=3D6>References </FONT><BR>[1] Friedman, J. H., Hastie, =
T. and=20
      Tibshirani, R., "Additive Logistic Regression: a Statistical View =
of=20
      Boosting." (Aug. 1998) <BR><BR>[2] A. Torralba, K. P. Murphy and =
W. T.=20
      Freeman. (2004). "Sharing features: efficient boosting procedures =
for=20
      multiclass object detection". Proceedings of the 2004 IEEE =
Computer=20
      Society Conference on Computer Vision and Pattern Recognition =
(CVPR). pp=20
      762- 769. </TD></TR></TBODY></TABLE></BODY></HTML>

------=_NextPart_000_0000_01C866A0.D0018460
Content-Type: image/gif
Content-Transfer-Encoding: base64
Content-Location: http://people.csail.mit.edu/torralba/shortCourseRLOC/boosting/boostingIcon.gif
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