can's profile - activity

Have you tried with different file formats?Like mp4,mkv or avi?

Well,then your best bet seems like training your own object detector.

Please provide more details.

Will the camera always be on these positions?

2) The negative grabber functions use a sliding window on negative images according to your model size (-w -h) and this is done each time when your image gets resized and it does this while maintaining the images original aspect ratio,otherwise you would end up with an artificial distortion which your original image didn't contain.

3) It is because your classifier starts to "learn" your dataset thus doing less mistakes on it.

what is the exact problem that you are facing?

can you provide details about positive and negative samples?

I thought so.

1080 is too expensive,is 980 or 980 Ti any good?

Hello,we all know when training a cascaded classifier your negative dataset shouldn't contain the object which you are trying to detect.My question is;what if a single image that contains your object gets overlooked and inserted into the negative dataset?How bad it would distort the final strong classifier?

as far as i know,in case of a tree-based classifier several features are combined together to form a decision tree with the -maxDepth parameter to create a single feature.

are you aiming for real-time?

can you provide more detail?

-numPos should be 0.9 * myVecFileCount

I have managed to tackle it with creating a custom class that inherits CascadeClassifier class than i used featureEvaluator->setImage on the image patch that i want to run my classifier on,after that i accessed the runAt() method which is protected and only way to access it is using a class wrapper.By supplying the correct model size and above instructions,it worked.

Hello everyone,how can i use the detectMultiScale method on a single specified location of specified size?I tried using it with same Size(x,y) parameters but it doesn't detect anything although i know that location and size contains my object of interest.

Hello,at cascadeclassifier.cpp file,method fillPassedSamples() there is this line

featureEvaluator->setImage(img, isPositive ? 1 : 0, i);

Can anybody provide a detailed description about what this line does?And after this line there is the below line,the point i fail to understand is;why isn't the predict() method not taking the img as argument?

if(predict(i) == 1.0F) { some code here }

neural networks

any links?

Hello,how can i build a simple motion indicator?For example; i want it to show a single value between let's say 1-100 where 1 would be no motion at all and 100 would be huge amounts of movement like the shaking of a handycam.

Well i kind of tried what you are asking with about 100 faces,the problem is;loading time of the model and the file size will increase rapidly thus making you wait alot for the model to load into memory.Another thing to take into account aswell is the sample count per face,i don't think you can give it a go with thousands of faces because of the file size of the model would grow beyond reason.

indeed it was and thanks for the proper one =D

Hello,how much does the traincascade application scales the negative images in the grabbing process at the start of each stage?I understand it uses a sliding window sized accordingly to your model parameters (-w,-h) on the images when searching for negative samples but i am curious about the scaling parameter,is it the default value 1.1?

I am suffering from the same problem,you can check my parameters below,the grabbing of negative samples decreases significantly when training the 4th stage (it can only grab around 8000 negatives in 24 hours),what can i do to make it grab faster?I have already tried offering a rich negative dataset with loads of images and it did not help much,instead when i supplied just 1 image in the -bg file the negative grabbing rate increased surprisingly and unexpectedly?!

opencv_traincascaded.exe -data classifier -vec samples.vec -bg negatives.txt -numPos 137 -numNeg 1250000 -numStages 4 -numThreads 4 -featureType LBP -w 50 -h 50 -minHitRate 1 -maxFalseAlarmRate 0.025

It should improve performance since it gets really annoying to train the deeper stages when you are aiming for a good classifier and by good i mean the overall false alarm rate should be like 4-5 negatives classified as positives in 7-8 million negatives.Another thing i have been thinking lately is that the traincascade application itself i guess has a failsafe to prevent running out of negative samples in the middle of the training,my guess is the program randomly rotates or changes brightness or contrast of the negative images to make the data set last long.What do you think about this? This should be taken into account as well.

I am indeed thinking of tackling it but right now i have a main project going on so i need to find a vacancy to make this going.OpenMP was the first thing that comes to mind,upon first look i immediately thought of;each thread needs to find numNeg/numThreads amount of negative windows but how come they can read the same data if you divide the negative data set?

Hello,i just wanted to ask that is there a way to parallelize the process of grabbing negative windows at the start of each stage in the traincascade application since that is the task that takes up alot of time when you are training the deeper stages.

Thanks.

training duration will depend on your dataset and parameters,it is true that you should not use all your positive samples in your .vec file since the algorithm will need more positive samples as you progress deeper in stages but i am not quite sure about the 1:2 ratio between positive and negative samples since i am not strictly following this rule in my attempts to train a cascaded classifier and get satisfied with the results. just test the end classifier with some test data and take notes;what does not satisfy you? if you have lots of false positives try increasing your numNeg parameter or reduce the maxFalseAlarm parameter or increase the numStages parameter,if you suffer from false negatives try increasing your positive samples.

Hello,the traincascade application and 32-bit operating systems won't do well together espacilly with HAAR features,i suggest using a 64-bit OS,when using traincascade with these parameters in my desktop pc (LBP features,2000 unique positve samples,450000 negative sub-windows,width 50,height 50) it is using almost 7 gigabytes of RAM.

Is there and implementation in OpenCV about this subject,if there isn't can anyone suggest a library implementing this algorithm?

Thanks

Yes,thanks for the answer.

you can crop the region of interest from the original image with

Mat myRoi = myOriginalImage(myRectangle).clone();

and then convert it to grayscale as usual with

cvtColor(myRoi,myRoi,CV_BGR2GRAY);