opencv 中的训练函数无法训练我的分类器

Question

I run these command

opencv_traincascade -data data/cascade -vec samples1.vec -bg negative/infofile.txt -numPos 231 -numNeg 100 -w 25 -h 15

in Centos system to train my classifier and it returns the follow

PARAMETERS:
cascadeDirName: data/cascade
vecFileName: samples1.vec
bgFileName: negative/infofile.txt
numPos: 231
numNeg: 100
numStages: 20
precalcValBufSize[Mb] : 256
precalcIdxBufSize[Mb] : 256
stageType: BOOST
featureType: HAAR
sampleWidth: 25
sampleHeight: 15
boostType: GAB
minHitRate: 0.995
maxFalseAlarmRate: 0.5
weightTrimRate: 0.95
maxDepth: 1
maxWeakCount: 100
mode: BASIC

===== TRAINING 0-stage =====
<BEGIN
POS count : consumed   231 : 231
Train dataset for temp stage can not be filled. Branch training terminated.
Cascade classifier can't be trained. Check the used training parameters.

Can anyone tell me what is the problem?:/

Answer 1

The number of negative samples is too small. You should increase it much larger such as 1000 by searching more background images on the web in which does not contain your detected object.

Answer 2

Train cascade doesn't pick your negative testdata. Please check the path specified in the negative/infofile.txt. One way to avoid this problem is using relative path in infofile.txt. Detailed information available here

Answer 3

One problem I encountered was the following: opencv_createsamples accepted relative paths, but opencv_traincascade did not accept relative paths.

To fix it, I created two different files for referencing the background images: bg_create.txt and bg_train.txt

The former having relative paths to the background images and the latter having absolute paths to the background images.