JunjieHPU's profile - activity

Apology about that Steven.

Hi, I just posted my script. Please take a look.

Hi guys,

I recently trained my own haar-feature xml file for detecting balloons. Then I wrote a script to parse the xml file and draw out all the haar features. However, some of them seem impossible like the following:

Are both of these two still considered as haar feature?

Thanks.

Here's my script.

I tried using the xml file with 0-12 stages, and the performance was pretty bad. I wonder if balloon is a good haar feature object and gonna re-test the performance using the newer xml file. BTW Steven, is it possible to see which haar feature it uses from the xml file?

Hi Steven,

1. Yes you are right about that. I should've said that the raw pictures' size are different;

2. My current version of OpenCV is 3.0.0; however, last night I stopped the program and re-ran the program started from stage 12, it took 11 hours to finish and now it's in stage 13 now. Isn't it weird?

3. There is a cascade.xml file in the folder. I'm gonna try it and see how it performs.

Thank you so much Steven.

Hi guys,

I'm trying to train my own xml file to detect balloons for my project. Here's my arguments when using traincascade in win7:

opencv_traincascade.exe -data classifier -vec samples.vec -bg neg.txt -numStages 20 -minHitRate 0.995 -maxFalseAlarmrRate 0.5 -numPos 610 -numNeg 3000 -w 60 -h 60 -mode ALL -precalcValBufSize 256 -precalcldxBufSize 256

FYI, the size of .vec file is about 3200. Sizes are different, most of them are 60x60.

Here's my question:

1. The program kind of stuck at stage 12 since it hasn't updated any information for more than 2 days
2. I tried to kill the program and generate a XML file with the existing 12 stages(0-11). However, the program just got killed after it displayed: Stages 0-11 are loaded.

Hope someone could explain. Thanks a lot.