lezan's profile - activity

Now is working like a charm. Thanks as always berak, your help is precious.

@dkurt that's a good point. I will try it.

Now is working like a charm. Thanks as always berak, you are help is precious.

Is it not enought to check the last layer of network? Perfect. Make sense. Thought could be a problem, it was not. Ok.

Hello, I playing with face and DNN but I cannot figure out of to solve an issue.

I am processing image 256x256. Using deploy.prototxt and res10_300x300_ssd_iter_140000.caffemodel (same one on dnn directory).

Some code.

cv::Mat faceROI;
cv::Mat image;

image = cv::imread(imagePath[imageId], CV_LOAD_IMAGE_COLOR);
cv::Mat imageDNNBlob = cv::dnn::blobFromImage(image, 1.0, cv::Size(300, 300), 
    Scalar(104.0, 177.0, 123.0), false, false);
netOpenCVDNN.setInput(imageDNNBlob, "data");
cv::Mat detection = netOpenCVDNN.forward("detection_out");
cv::Mat faces(detection.size[2], detection.size[3], CV_32F, detection.ptr<float>());
for (int i = 0; i < faces.rows; i++)
{
    float confidence = faces.at<float>(i, 2);
    if (confidence > 0.99)
    {
        int xLeftBottom = static_cast<int>(faces.at<float>(i, 3) * image.cols);
        int yLeftBottom = static_cast<int>(faces.at<float>(i, 4) * image.rows);
        int xRightTop = static_cast<int>(faces.at<float>(i, 5) * image.cols);
        int yRightTop = static_cast<int>(faces.at<float>(i, 6) * image.rows);

        cv::Rect faceRect((int)xLeftBottom, (int)yLeftBottom, 
                    (int)(xRightTop - xLeftBottom), (int)(yRightTop - yLeftBottom));
    faceROI = cv::Mat(image, faceRect);
         }
 }

Nothing too exotic, I just write down what I found in resnet_ssd_face.cpp. When I try to extract ROI from image with faceROI = cv::Mat(image, faceRect) I get an error on wrong dimensions with faceRect, in fact (with a particular image) I get 257 as dimension (height). faces.at<float>(i, 6)return a float >1.

What I miss? Can some help to figure out?

I have also some questions about this example:

1. netOpenCVDNN.forward return a Mat, where size[2] is the number of object found, size[3] numbers of property of each object? Am I right? Where can I find more info about what forward return? (Already checked here and here. I think it is related to the layer "detection_out" of prototxt, but I can not get it).
2. Mat facesis a matrix with all faces found, right? Where each rows is a face detected and each rows (face) have some property (cols), right? So faces.at<float>(i, 2) is the confidence of i-th face and from 3 to 4 are dimensions of face. What position 0 and 1 contains?
3. Why cv::Mat imageDNNBlob have a numbers of rows and cols like -1?
4. Last one: I am using image of 256x256 dimension. Input layer of dnn use 300x300 as dimension. What is the right solution? Resize image? Change input layer? Is cv::Size(300, 300) right in blobFromImage?

Thanks in advance.

DNN opencv with SSD resnet return wrong face dimension Hello, I playing with face and DNN but I cannot figure out of to

DNN opencv with SSD resnet return wrong face dimension Hello, I playing with face and DNN but I cannot figure out of to

As reported here Visual Studio 15.6 is the last one supported for CUDA 9.2. If you want to run (compile etc) CUDA with y

It does not make much sense to be wrong?

@berak Is there no way to use a grayscale image? You suggest to use color images just for dim=3 channels in prototxt? Sw

What version of Visual Studio are you using? 15.6 is the last one supported by CUDA 9.2. P.S.: next time do not put you

What version of Visual Studio are you using? 15.6 is the last one supported by CUDA 9.2. P.S.: next time do not put you

What version of Visual Studio are you using? 15.6 is the last one supported by CUDA 9.2.

Is there something that can discriminate between them? For example, image name. However this is not a question related t

Is there something that can discriminate between them? For example, image name. However this is a question related to Op

Is there something that can discriminate between them? For example, image name.

Add some code and image example.

You can try in this way: Apply threholding because you need a binary image. Apply morphological operator: first Open,

You can try in this way: Apply threholding because you need a binary image. Apply morphological operator: first Open,

I get it, but how do you install opencv on your machine? Do you compile it or get binary from site?

How you compile OpenCV? With VC? MinGW?

Have you checked documentaion here https://docs.opencv.org/3.4.1/de/d25/imgproc_color_conversions.html?

https://docs.opencv.org/3.4.1/de/d25/imgproc_color_conversions.html checked?

Well, without any code is difficult to answer, but I guess that OpenCV is not correcly installed, at least in release. H

Have you try to switch from Debug mode to Release mode? Do you add AVX, AVX2 or something else?

After some months, I start again to do some optimization with my code, but I forgot somethink, I think.

I need to perform Bag of Words to clusterize features extracted from images. Let's see some code.

Extract features and put them into a vector.

cv::Mat featuresVector;
for (int i = 0; i < numberImages; ++i) // <- first features extraction
{
    cv::Mat featuresExtracted = runExtractFeature(face, featuresExtractionAlgorithm);
    featuresVector.push_back(featuresExtracted);
}

Them, I want to clusterize them with BOWKMeansTrainer.

cv::BOWKMeansTrainer bowTrainer(dictionarySize, termCriteria, retries, centersFlags);
bowTrainer.add(featuresVector);
cv::Mat dictionary = bowTrainer.cluster();

Then, preare for bag of words in this way

cv::Ptr<cv::DescriptorMatcher> matcher = cv::FlannBasedMatcher::create();
cv::Ptr<cv::DescriptorExtractor> extractor = cv::xfeatures2d::SiftDescriptorExtractor::create(); // <-
cv::BOWImgDescriptorExtractor bowDE(extractor, matcher);
bowDE.setVocabulary(dictionary);

Now I can start bag of words

for (int i = 0; i < numberImages; ++i)
{
    cv::Mat face = faceMatVector[i]; // <- contains image read with imread.
    std::vector<cv::KeyPoint> keypoints;
    detector->detect(face, keypoints); <- // second one features extraction
    cv::Mat bowDescriptors;
    bowDE.compute(face, keypoints, bowDescriptors);
}

As you can see, in this way, I perform features extraction from image two time: first one with the first loop, and the second one with the last loop (see comments on code), because I need descriptors to clusterize with BOWKMeansTrainer, but I need keypoints to calculate bowDescriptors with BOWImgDescriptorExtractor (matching and so on).

My question is: Is this necessary or can I avoid that? I failed something? Can I take from keypoints from somewhere in the last loop without re-detect them? Can I just save keypoints detected in the first loop and then re-use them in the last loop to computer BOWImgDescriptorExtractor?

Thanks for your answer.

So can I use descriptors saved into featuresVector? Seems awesome, did not see this overloading.

BOWKMeansTrainer and features extraction After some months, I start again to do some optimization with my code, but I fo

I tried with LINEAR, but result is not good (around 40%). About the last question: I did no try, but - as a profane - se

Nope, I only try to go with my own solution. I thought you were spurring me in this direction :D. By the way, now just f

Hey @berak, have you seen the last edit of question? I tried to follow your suggestion and for now I am working with dat

SVM predict on OpenCV: how can I extract the same number of features I am play with OpenCV and SVM to make a classifier

SVM predict on OpenCV: how can I extract the same number of features I am play with OpenCV and SVM to make a classifier

Back from work, I have two questions, if you can answer: I need to re-use centers for the unseen images, right? I do n

Back from work, I have two questions, if you can answer: I need to re-use centers for the unseen images, right? I do n

I missed the last comment and I am waiting here from 6 hours, auch. Anyway, I cannot understand why my approach is wrong

I read it 10 times but I am bit confused. In particular, I am refering to the third point. Okay, I get it. Get it too.