{"id":106,"date":"2009-11-10T16:59:44","date_gmt":"2009-11-10T06:59:44","guid":{"rendered":"http:\/\/www.computer-vision-software.com\/blog\/?p=106"},"modified":"2009-11-30T13:01:11","modified_gmt":"2009-11-30T03:01:11","slug":"faq-opencv-haartraining","status":"publish","type":"post","link":"http:\/\/www.computer-vision-software.com\/blog\/2009\/11\/faq-opencv-haartraining\/","title":{"rendered":"FAQ: OpenCV Haartraining"},"content":{"rendered":"

Hi All, before posting your question, please look at this FAQ<\/span> carefully!<\/span> Also you can read OpenCV haartraining article<\/a>.\u00a0 If you are sure, there is no answer to your question, feel free to post comment.\u00a0 Also please, put comments about improvement of this post.\u00a0 This post will be updated, if needed.<\/p>\n

<\/h1>\n

Positive images<\/span><\/em><\/h2>\n


\n<\/span><\/em><\/p>\n

Why positive images are named so?<\/em><\/h3>\n

Because a positive image contains the target object which you want machine to detect. Unlike them, a negative image doesn’t contain such target objects.<\/p>\n

What’s vec file in OpenCV haartraining?<\/em><\/h3>\n

During haartraining positive samples should have the same width and height as you define in command “-w -h size”.\u00a0 So original positive images are resized\u00a0 and packed as thumbs to vec file. Vec file has header: number of positive samples, width, height and contain positive thumbs in body.<\/p>\n

Is it possible to merge vec files?<\/em><\/em><\/h3>\n

Yes, use Google, there are free tools, written by OpenCV’s community.<\/p>\n

I have\u00a0 positive images, how create vec file of positive samples?<\/em><\/h3>\n

There is tool in C:\\Program Files\\OpenCV\\apps\\HaarTraining\\src createsamples.cpp.\u00a0 Usage:<\/p>\n

createsamples -info positive_description.txt -vec samples.vec -w 20 -h 20<\/p>\n

What’s positive description file?<\/em><\/h3>\n

The matter is that, on each positive image, there can be several objects. They have bounding rectangles: x,y, width, height.\u00a0 So you can write such description info of image:<\/p>\n

positive_image_name\u00a0 num_of_objects x y width height x y width height …<\/p>\n

Text file, which contains such info about positive images is called description file. So during vec file generation,\u00a0 really objects are packed<\/span>, but not whole image. Essentially vec file is needed to speed up machine learning.<\/p>\n

Do I always need description file, even if I have only one object on a image?<\/em><\/h3>\n

Yes, with createsamples you need description file.\u00a0 If you have only one object, it’s bounding rectangle may be bounding rectangle of whole image. If you want, write your own tool for vec file generation =)<\/p>\n

Should lightning conditions and background be various on positive images?<\/em><\/h3>\n

Yes, it’s very important<\/span>. On each positive image, beside object, there is background. Try to fill this background with random noise, avoid constant background.<\/p>\n

How much background should be on positive image?<\/em><\/h3>\n

If you have much background pixels on your positive images in comparison with object’s pixels – it’s bad since the haartraining could remember the background as feature of positive image.<\/p>\n

If you don’t have background pixels at all – it’s also bad. There should be small background frame on positive image<\/p>\n

Should all original positive images have the same size?<\/em><\/h3>\n

No, original images can have any size.\u00a0 But it’s important that width, height of this rectangle have the same aspect ratio as -w -h.<\/p>\n

What’ s\u00a0 -w and -h should I put in createsamples? Should it be always square? <\/em><\/em><\/h3>\n

You can put any value to -w and -h depend on aspect ratio of the target object which you want to detect.\u00a0 But objects of smaller size will not be detected<\/span>! For faces, commonly used values are 24×24, 20×20. But you may use 24×20, 20×24, etc.<\/p>\n

Errors during vec file generation: <\/em>Incorrect size of input array, 0 kb vec file,<\/h3>\n

-First check you description file: positive_image_name should be absolute path name<\/span> without spaces like “C:\\content\\image.jpg” not “C:\\con tent\\image.jpg” or relative path name<\/span>.<\/p>\n

-Avoid empty lines in description file<\/p>\n

-Resolution of original positive image file should be not less, then -w -h parameters you put.<\/p>\n

-Check that positive images are available in your file systems and not corrupted.<\/p>\n

-There can be unsupported formats. Jpeg, Bmp, PPM are supported!<\/p>\n

Example of vec file generation!<\/em><\/h3>\n

Let’s working directory be C:\\haartraining. In it there is createsamples.exe. There is folder<\/p>\n

C:\\haartraining\\positives. So create description file positive_desc.txt.<\/p>\n

positives\\image1.jpg 1 10 10 20 20<\/p>\n

positives\\image2.jpg 2 30 30 50 50 60 60 70 70<\/p>\n

or<\/p>\n

C:\\haartraining\\positives\\image1.jpg 1 10 10 20 20<\/p>\n

C:\\haartraining\\positives\\image2.jpg 2 30 30 50 50 60 60 70 70<\/p>\n

You should avoid empty lines and empty space in image’s path<\/p>\n

createsamples -info positive_desc.txt -vec samples.vec -w 20 -h 20<\/span><\/span><\/p>\n


\n<\/strong><\/em><\/span><\/h2>\n

Negative images<\/strong><\/span><\/em><\/h2>\n


\n<\/strong><\/span><\/em><\/p>\n

What negative images should I take?<\/em><\/h3>\n

You can use any image of OpenCV supported formats, which does not contain target objects (which are present on positive images). But they should be various<\/span> – it’s important! Good enough database is here<\/a><\/p>\n

Should negative images have the same size?<\/em><\/h3>\n

No. But the size should not be less than -w -h, which were put during vec file generation.<\/p>\n

What’s description file for negative image? <\/em><\/h3>\n

It’s just text file, often called negative.dat, which contains full path to negative images like:<\/p>\n

image_name1.jpg<\/p>\n

image_name2.jpg<\/p>\n

Avoid empty lines in it.<\/p>\n

How many negative\/positive image should I take?<\/em><\/h3>\n

It depends on your task.\u00a0 For real cascades there should be about 1000 positive images and 2000 negative images e.g.<\/p>\n

Good enough proportion is\u00a0 positive:negative = 1:2, but it’s not hard rule! I would recommend first to use small number of samples, generate cascade, test it, then enlarge number of samples.<\/p>\n

Launch haartraining.exe (OpenCV\\apps\\HaarTraining\\src<\/strong><\/em><\/span>)<\/span><\/strong><\/span><\/em><\/strong><\/em><\/h2>\n


\n<\/span><\/strong><\/span><\/em><\/strong><\/em><\/p>\n

Example of launching<\/em><\/em><\/h3>\n

Working directory is C:\\haartraining with haartraining.exe tool and samples.vec file.<\/p>\n

Let’s negative images are in C:\\haartraining\\negative, in this case negative.dat should be like this:<\/p>\n

negative\\neg1.jpg<\/p>\n

negative\\neg2.jpg<\/p>\n

…<\/p>\n

So in C:\\haartraining launch this:\u00a0 haartraining<\/span> -data haarcascade -vec samples.vec -bg negatives.dat -nstages 20\u00a0 -minhitrate 0.999 -maxfalsealarm 0.5 -npos 1000 -nneg 2000 -w 20 -h 20 -nonsym -mem 1024<\/p>\n