Poyan's profile - activity

Is it possible to get a prediction's probability along with the prediction when using an SVM in OpenCV?

I found some threads discussing this, but they are all from 2-3 years ago.

http://answers.opencv.org/question/14...

http://stackoverflow.com/questions/16...

Any advances on this?

I'm using DenseFeatureDetector and want to return only a certain number of keypoints (or as close to that amount as possible). I know I can tweak the parameters listed in the documentation (http://docs.opencv.org/modules/featur...) and try to calculate how many keypoints will be generated, but I was wondering if there's a simpler way.

With the other feature detectors (SIFT, FAST etc) I can sort based on the response and only save the n best features. Since the DenseFeatureDetector only lays out the keypoints in a gridlike fashion, the points don't have a "response" value AFAIK.

I could just take the first n features, but that would probably only cover the first portion of the page. Better would be to save every k keypoint so that it is still in a gridlike fashion but just more sparsely laid out.

tl; dr: What's the best way to return only n features when using DenseFeatureDetector?

Guanta, I will read up on AKAZE and perhaps update the test!

It is only descriptor extraction. I will probably do a comparison of matching times within the next few weeks. My main purpose is image recognition though, and not feature tracking.

Is this really still active?

By the way, perhaps you have some insight on a related question? http://answers.opencv.org/question/58...

Interesting find Guanta, I was not aware an implementation for k-majority for binary data existed, I actually implemented it myself last week. Perhaps it isn't common yet because it hasn't been readily available to do the clustering.

I'm currently looking into if it's feasible to do image recognition with binary feature descriptors, I will report back with my results.

My impression from reading a number of associated papers is that binary feature descriptors would be an order of magnitude faster (or more!) than for instance SIFT.

However, my results in a simple benchmark do not show this. All images + keypoints were pre-cached before running the benchmarks, so it's not that.

 DESCRIPTOR              MS        FEATURES          IMAGES      MS/FEATURE
      BRIEF             612          263120             230        0.002326
      BRISK             964          263120             230        0.003664
      FREAK             564          263120             230        0.002144
        ORB             476          247250             230        0.001925
       SIFT            1047          247250             230        0.004235

What's the reason for the poor results? Is it because of the implementations in OpenCV? Or am I missing something else?

EDIT: For instance, in the FREAK-paper SIFT-description took 2.5 ms per key point vs 0.018 per key point for FREAK, that is, the latter is about 138 times faster to compute.

@Guanta, what's the basis behind the comment "BRISK / FREAK / ORB were all intended for a fast matching for tracking and not for an image classification problem."?

No update on this? I realized this as well and posted a more general question here; http://answers.opencv.org/question/57...

I was under the impression that since the Java bindings were autogenerated from the C++ library, everything would be available. However, there seems to be quite a bit of things missing, for example FLANN and BOW-related stuff.

What is the best approach for dealing with this?

My questions are

1. Why is stuff missing?
2. Is it possible to recompile the library but while generating more of the bindings?
3. Is there any good guide on how one can "easily" use JNI to access the missing portions?

Also, I think there should be a warning on the OpenCV Java-download page that the included bindings are not complete.

I have the following snippet which successfully gets a matrix element's value(s).

However, the values I try to put back, never actually update the matrix object.

Any ideas?

        float[] floats = new float[imgDescriptor.channels()];
        imgDescriptor.get(1, trainIdx, floats);
        // Just in case, we loop through the whole array
        for (int j = 0; j < floats.length; j++) {
            floats[j] = floats[j] + 1.0f;
        }
        imgDescriptor.put(1, trainIdx, floats);

Any advances on this?