Comments on: FAQ: OpenCV Haartraining

By: Guest

Guest — Thu, 26 May 2011 08:25:53 +0000

Does anybody know why I get an error while running haartraining.exe?

Error: Invalid background description file

By: Nastia

Nastia — Wed, 25 May 2011 14:39:07 +0000

Hello, I’d like to get your advice and maybe you can help me to understand what mistake I make. I stuck with Haartraining function. I have generated delnas.vec file. My decription file looks like:

C:\Nuotraukos\Delnas\Neigiamos\nuotr265.jpg 1 170 100 200 200
C:\Nuotraukos\Delnas\Neigiamos\nuotr266.jpg 1 150 80 250 280

The main my work directory is C:\Nuotraukos. I moved there all neccessary files: opencv_haartraining.exe, delnas.vec, duomenys.txt…
After I run the command:
C:\Nuotraukos>C:\OpenCV2.2\bin\opencv_haartraining.exe -data C:\Nuotraukos -vec
delnas.vec -bg duomenys.txt -npos 519 nneg 1081 -nstages 20 -mem 1500 -mode ALL
-w 20 -h 20

(for haartraining.exe file I had to specify the whole dirtectory to Opencv2.2: C:\OpenCV2.2\bin\opencv_haartraining.exe, otherwise I get an error that msvcp100.dll is missing despite the fact that I moved opencv_haartraining.exe to C:\Nuotraukos). So after I get an error:

Height: 20
Applied boosting algorithm: GAB
Error (valid only for Discrete and Real AdaBoost): misclass
Max number of splits in tree cascade: 0
Min number of positive samples per cluster: 500
Required leaf false alarm rate: 9.53674e-007

Tree Classifier
Stage
+—+
| 0|
+—+

Number of features used : 67074

Parent node: NULL

*** 1 cluster ***
POS: 519 519 1.000000
Invalid background description file.

Do you know what it depends on? Please reply back to me asap directly to my e-mail: anastasija.kac@gmail.com

By: Bui Quang Hien

Bui Quang Hien — Wed, 25 May 2011 12:35:49 +0000

Hi,
I want to detect character of license plate using adaboost + haarlike features. createsamples -info pos.txt -vec pos.vec -w 20 -h 20. I want to detect object with small size. I write file pos.txt: pos\1.jpg 1 0 0 20 20. Should I crop all image of object into square (410×410,900×900)? After that, I resize all them into 20×20 and convert into gray image.
Please give me some advice!

By: Igor

Igor — Thu, 19 May 2011 18:45:29 +0000

OK now I set -nneg 1 and it works, but when i put -nneg 10 i still doesn’t work…Now I have about 500 negative images in negs.txt…
What is the problem….Why can’t I set -nneg at some normal value

By: Igor

Igor — Wed, 18 May 2011 14:19:59 +0000

Hi,
I have a problem running opencv_haatraining…I have 873 positive images and about 200 negative images….All negative images are big usually 1280*1024…(-nneg is about 1500)
When I run opencv_haartraining after a while it just stops…It works normal for few hours and then it stops at this line

Parent node: X

*** 1 cluster ***
POS: 864 875 0.994942

There is to percentage shown as it usually is, and line NEG: is not showing…
I saw at some previous posts that you suggested to try different or bigger negatives…I tried that all but it still doesn’t work…
Can anyone please answer this question it is really important…
Thank you…

By: michaela

michaela — Sun, 08 May 2011 02:09:46 +0000

I am making a Hardware implementation of Viola jones and it’s very important to know the exact way to use the threshold value.

I read that the variance normalization is suggested by Viola and Jones as a mean of reducing that effect of different lighting conditions.
So, for to realize an implementation of Viola and Jones, the threshold value given in the .xml is multiplied for the standard desviation, according to the code performance.cpp.
So i perceive is — Calculate variance – calculate Standard deviation from it – and multiply it with the Threshold and then compare it with the normal feature value:

if(feature value < threshold*Standard desviation) then
left value given in the .xml
else
right value given in the .xml

I'm not sure….Someone can explain it better? and Why the threshold is multiplied for the standard desviation??

By: Thuy Dinh

Thuy Dinh — Mon, 25 Apr 2011 18:51:49 +0000

Hello!
I don’t know how normalized image and create the positive description file?
Please, help me to solve this problem? I am really very grateful

Thanks.

By: Olu

Olu — Wed, 13 Apr 2011 19:30:23 +0000

What is the format of you negative images?

By: Alexander Gavrik

Alexander Gavrik — Wed, 13 Apr 2011 02:17:51 +0000

Hi Jack, as mentioned in article above in section “Is it possible to detect rotated faces?”, you can detect face at 90 degree (as a separate cascade). However you can’t detect face in all orientations with single cascade using this method.

By: Olu

Olu — Mon, 11 Apr 2011 12:34:22 +0000

Hello,

I recently said this video of the work of Zdenek Kalal (http://info.ee.surrey.ac.uk/Personal/Z.Kalal/tld.html ). He created TLD which is able to learn and track an object at the same time. For object tracking in OpenCV, is there a way of putting a boundary rectangle around the object of interest and automatically track and generate the description file?