OpenCV: good learning outcome, but cascading classifier

Question

Very new to OpenCV, and I try my best to train a harass classifier that can detect dog images from a side overlay. I used this tutorial as a guide. The author suggests that a relatively effective classifier can be trained using a surprisingly small number of model images. According to his instructions, I collected 40 positive and 600 negative, and then used the script to generate many more samples as .vec files. The training took about a week and a half to 20 stages with the following parameters:

<?xml version="1.0"?>
<opencv_storage>
<params>
  <stageType>BOOST</stageType>
  <featureType>HAAR</featureType>
  <height>64</height>
  <width>80</width>
  <stageParams>
    <boostType>GAB</boostType>
    <minHitRate>9.9900001287460327e-01</minHitRate>
    <maxFalseAlarm>5.0000000000000000e-01</maxFalseAlarm>
    <weightTrimRate>9.4999999999999996e-01</weightTrimRate>
    <maxDepth>1</maxDepth>
    <maxWeakCount>100</maxWeakCount></stageParams>
  <featureParams>
    <maxCatCount>0</maxCatCount>
    <featSize>1</featSize>
    <mode>ALL</mode></featureParams></params>
</opencv_storage>

0,000579, , 0,0579% , .. , . , 99,942% . , , .xml , .

, ( , ):

( , , . , , , ...)

, .

, , : ? ? , , ?

.

Answer 1

, , :

• ( ) . .
• overfited: ( 1-6). , ( ), . , ( , ).

. , HOG, LBP. featureType HOG LBP.

. , ( ), ( > 500), ( ).

(min-, , minSize maxSize).