StevenPuttemans's profile - activity

For the purpose of algorithm comparison I want to compare the detection result of my cascade classifier trained using the viola & jones framework versus an SVM HOG classifier I have trained.

However, the VJ framework has no problem with an uneven ratio for positives and negatives, for example 500 positives and 100 negatives (don't tell me this is wrong please, my whole PhD is about proving that context awareness can reduce the negative numbers effectively. .

If I would like to convert this to SVM classification however, I am reading at many blogposts that using CvSVM should be done with an equal number of positive and negative training samples. If not I should add a bias, but however, I cannot find an example that does exactly that.

Could anyone give me some directions on how I should apply this? Is this even possible within the OpenCV framework?

Kind regards,

Steven

EDIT 1: I have noticed people suggest using libSVM or SVMlight for training the model. However, I would like to stick to openCV only, which should actually be possible.

Is it possible to do?

As of know I am using the following to pass an image to the callback:

static void onMouse( int event, int x, int y, int, void* param)
{
    // Mount back the parameters
    Mat* ptrImage = (Mat*)param;

    //What to do when left button is pressed
    if( event != EVENT_LBUTTONDOWN ){
        // Draw the point on the image here!
        return;
    }
}

namedWindow("draw mask", WINDOW_AUTOSIZE);
setMouseCallback("draw mask", onMouse, (void*)&work_image);

But I would like to add a second pointer to a vector of points to keep storing them also, just don't know how to add multiple parameters. Anyone has any idea without creating global variables?

I see many topics appearing about cascade training for object detection. However, what I do not seem to find are discussions about how fluently training should actually be.

I am training a LBP object detector using the boosted cascade of weak classifiers, done by using the train_cascade algorithm. The detector is used for detecting cars in aerial imagery.

I have used 200 positive examples and 1000 negative examples. I asked to train 35 stages, which the trainer is actually doing, but when I check how fast the algorithm is processing the negative samples, I see weird results.

At first stages processing goes very fast. At stage 19 however, processing goes about a sample each 5 - 10 minutes. Just wondering, since LBP has been reported as beïng trainable in matter of a day, where HAAR needs a week, how fast training goes with others.

Also, I have the idea that assigning more memory to the process, by increasing the precalcValBufSize and precalcIdxBufSize parameter to 2048 MB each, doesn't seem to influence the training more, than actually assigning 500 MB to each. Any suggestions there?

UPDATE

I have done multiple tests on this, still not being able to define a relation between LBP training time and the algorithm itself. Still open for suggestions!

I got a function

void on_mouse(int event, int x, int y, int flag, void *param)
{
   //mouse functionality here
}

Then the function detectPositives() does the following command inside

cvSetMouseCallback(window_name, on_mouse, NULL);

And it doesn't seem to work using the function name as function pointer, which works perfectly in a CLI application, but not in this WinForm application header file.

Error message = error C3867: 'ObjectDetectionGUI::Form1::on_mouse': function call 
missing argument list; use '&ObjectDetectionGUI::Form1::on_mouse' to create a pointer 
to member

Anyone could help?

This is not an error! Given the samples presented and the settings of the training, your cascade has reached its desired

You are hitting the nail there ... since it is indeed the max allowed FArate, it should be lower to be more strict, and

Exactly my point, on 2 OS systems, even reading from the exact same source, could be given different values.

The documentation is indeed not that clear. We might want to propose a pull request with a chance on that level.

Its just strange to copy paste someones question. If you have the same issue, then simply add a comment beneath the ques

My guess is the internal translation of the indexes is somehow broken ... could you maybe open up an issue at github wit

Hahaha I am totally unrelated to nvidia :D Never did any work for them or with them. Deepfakes are not the solution howe

@NickKao, dont worry. I deleted them. Seems to be a system fault :)

Then my guess is this issue is due to your Qt environment are wrongly configured and the issue is not within OpenCV itse

did not know that :D thanks for giving a heads up

can you add between each read() operation a wait? I know that when grabbing frames in a loop, in C++ I always add a wait

But it tells us what you are trying to change. You cannot expect us to dig into the code to figure it out. Are you sure

On top of that, there is this book opencv 3 blueprints, which has a very detailed discussion on the process as well as a

You cannot just transfer dlls from one machine to another without having an exact copy. It is very tricky. I would urge

That is indeed weird. @markcowell23, be warned that it is not normal simply copy pasting others questions. If this keeps

@holger, there is no way you can mimic facial microexpressions over time inherently specific to the person in question.

The duplicate posts have been deleted ... you posted 14 times the same question. In the future, please pay attention to

As you can see in the moments tutorial you need a contour, which is a vector of point elements as input format. My guess

Keep in mind that the facemark code was developed during GSoC2017 and is still since then in contrib module because of t

if you really want quick help on DL issues, open up a full report on github repo and @dkurt will jump in quite suddenly

Prebuilt OpenCV has no OpenCL support I think ... but anyway, its always better to select all options yourself and build

So basically what happened is that your source systems has another video capturing backend installed which is grabbed be

Its somewhat out of my expertise and not really sure how to help you further...

@holger, your comment makes absolutely no sense :D

yep that seems correct, and indeed I guess you will need to build OpenCV first before your top project and not during th

yep that seems correct

Can you please use the correct flags instead of integer values, as described here?

Can you please use the correct flags instead of integer values, as described (here)[https://docs.opencv.org/3.1.0/d8/dfe

Well with mixing I mean alpha blending. Your border is so huge it has to inpaint pixels from far away. Also the borders

Ow your approach is sow flawed and for the following reasons haartraining is an old tool, it was replaced by traincasc

That error is not due to OpenCV code but an output error of your realsense sensor, as described here ...

You are retrieving the watershed regions as markers. These are basically points that have a specific ID. You can add all

Can you check if adding an explicit clone after the frame grabbing, so within the loop, solves the issue? Like a hard lo

The only time I get such link errors is when I try static linking and order of libraries is incorrect. Is your library l

BTW, your approach is broken in the beginning with the matcher. It will never work since your calibration pattern covers

Have a look at the coin segmentation example using watershed. If you calculate the distance transforms, than threshold i

Have a look at the coin segmentation example using watershed. If you calculate the distance transforms, than threshold i

_NORMED only means range normalized values in the case of the opencv cross correlation implementation. If I have a look

It seems, using the knowledge of this SO issue that data collected from source on 2 OS systems could have slight differe

If you really want people to dig into your issue, create a code sample to illustrate your problem and add it to the post

There is a whole range of books on OpenCV for android and some of them target specific platforms like Android. I suggest

@Eduardo, that is what I mean, but the problem is that he is running on Windows, on which collecting that info is not as