29 - Key points, detectors and descriptors in openCV

Поділитися
Вставка
  • Опубліковано 13 січ 2025

КОМЕНТАРІ • 61

  • @RichieChuh
    @RichieChuh 4 роки тому +6

    The content you presente is so briliant!

  • @dheerendrabhandari751
    @dheerendrabhandari751 Рік тому +1

    Thanks for the structured session on image proccesing 🤘

  • @TinneMertens-lo9rl
    @TinneMertens-lo9rl 8 місяців тому

    Thanks!

  • @alexandrakhlivnenko6960
    @alexandrakhlivnenko6960 Місяць тому

    really loved your explanation, thanks.

  • @rakshitjoshi823
    @rakshitjoshi823 2 роки тому +1

    At 19:30 where you were explaining about FAST detector algorithm. I think that the function 'FastFeatureDetector_create(50)' is not taking 50 points as when image is generated the numbers are quite low. To back this, you can see when you change value from 50 to 20, the number of corner points in new image are far more than the previous case. This means that 50 is not about 50 points.
    Let me know if I am wrong and what truly create is doing over there?

  • @osamamohamed5609
    @osamamohamed5609 5 років тому +1

    I need to know more about the difference between the key points and descriptors 😁
    And I was in need to use SIFT

  • @Spayke201
    @Spayke201 5 років тому +1

    Very nice tutorial! It helps me a lot! Thank you.

  • @yehorsemeniaka6951
    @yehorsemeniaka6951 5 років тому +2

    Very good tutorial, thank you!

  • @akmmuhitulislam2629
    @akmmuhitulislam2629 3 роки тому +1

    This was so helpful. Thank you very much.

  • @rverm1000
    @rverm1000 11 місяців тому

    how would you go about figuring if star images have a moving object in it?

  • @akashravi9930
    @akashravi9930 2 роки тому

    Future here. Have you done on SWIF or SURF later after that.

  • @paguevara
    @paguevara 3 роки тому

    Do you have the LESH descriptor?

  • @madhavimehta6010
    @madhavimehta6010 3 роки тому

    it's very insightful. Thanks for your effort

  • @abhishekg4147
    @abhishekg4147 Рік тому

    can we use custom matching points? like if we can manually add what all the required features we need to do template matching?

  • @AI_For_Scientists
    @AI_For_Scientists 4 роки тому

    Excellent video! Thank you.

  • @maudiakhumaidasavitri205
    @maudiakhumaidasavitri205 4 роки тому

    How can i get orientation, scale and position information of keypoints?

  • @kollurisharada5102
    @kollurisharada5102 2 роки тому

    Sreeni Garu , Thank You so much for wonderful session, May I know how to communicate you🙏🙏🙏

  • @gupta942
    @gupta942 4 роки тому

    Amazing explanation Sir 👌👌👌

  • @bellooluwaseyi4193
    @bellooluwaseyi4193 4 роки тому

    Excellent! Please can you do a video on how to use these methods for classification/recognition, like video in 176

  • @samardas1365
    @samardas1365 4 роки тому

    thank you , is there any tutorial on gist descriptor?

  • @hasanjameel2148
    @hasanjameel2148 3 роки тому

    how i can "brisk detector" to detect points in npyhon please

  • @ElDiablo123
    @ElDiablo123 3 роки тому

    Can you do about fragile telomeres?

  • @deepanshuvishwakarma316
    @deepanshuvishwakarma316 5 років тому +1

    Amazing tutorial.. thanks a lot

  • @sanisalisu7201
    @sanisalisu7201 Рік тому

    Hello there, this is very helpful. my code is not using cv2.imshow. is reporting the cv2.imshow is not in use in Colab. when I use display, the image did not show on the window

  • @caiyu538
    @caiyu538 3 роки тому

    great lectures

  • @alekamelese98
    @alekamelese98 3 роки тому

    Thank you for your explanation.. How to combine CNN features with HOG/SIFT?

  • @michellesainos3443
    @michellesainos3443 4 роки тому

    Great tutorial! thank you very much

  • @steliosneophytides7060
    @steliosneophytides7060 4 роки тому

    Hello, do you know how to find the key points on video's frames?

    • @DigitalSreeni
      @DigitalSreeni  4 роки тому +1

      Sorry, never tried it on videos but I don;t see why it would be any different. You just load a video as (time) series of images so I believe this procedure should work.

    • @steliosneophytides7060
      @steliosneophytides7060 4 роки тому

      @@DigitalSreeni Yeah i tried It but is not working as in an individual image.
      If i find something, i will come up with Informations

  • @lakpatamang2866
    @lakpatamang2866 4 роки тому

    How can i match two images using those descriptors? Suppose i want the matching score between two images on scale of 0-1. Can i use those descriptors to do the same?

  • @janiksmejkal1474
    @janiksmejkal1474 4 роки тому +2

    cv2.FastFeatureDetector_create(50) --> parameter '50' is for threshold, not for number of key points. Otherwise, brilliant series!!!!!

    • @DigitalSreeni
      @DigitalSreeni  4 роки тому +2

      Thanks for correcting me, really appreciate.

    • @ranati2000
      @ranati2000 2 роки тому

      Hi, is it related to limiting number of features so that every image can have same number of features so that we can use svm or any other traditional ml algos on it ?

  • @siddharthsvnit
    @siddharthsvnit 2 роки тому

    i think license for SIFT has expired in 2020 and its implementation is available in OpenCV main branch

    • @DigitalSreeni
      @DigitalSreeni  2 роки тому

      Yes, you are right. I checked it on opencv 4.6.0 and SIFT is available. Thanks for the note.

  • @sarkar0-z7z
    @sarkar0-z7z 4 роки тому

    Sir why didn't you used this method to convert the image to grayscale "img = cv2.imread("images/grains.jpg",0) " like the way you did in last tutorial rather you used "gray=cv2.cvtColor() "?
    Is their any difference between the two methods ? And if there is what is that?

    • @DigitalSreeni
      @DigitalSreeni  4 роки тому +2

      No difference between both approaches. No specific reason why I chose one over the other method. Normally I like to import images as color (RGB) and then convert them to grey scale. That way my color image is available in case I need it for some operations. Also, I sometimes apply an image processing function to an RGB image by converting it to HSV space or applying it to each of the R, G, B channels. In summary, import images as greyscale if you know you do not need all channels.

  • @marcindyba4774
    @marcindyba4774 Рік тому

    Dear Sreeni Garu,
    First of all thank you for your time and very professional explanations and tutorials.
    I struggling to save kp1 and des1 after orb.detectAndCompute I am going to use in another project.
    (kp1,des1=orb.detectAndCompute(img1,None))
    I have seen your next tutorial about cvs save , but I can not do it for this kp and des.
    Please help when you can , I would appreciate.

  • @NotYourDud3
    @NotYourDud3 2 роки тому

    Can somebody simplify "descriptor" and "key point" for me, please? Still struggling to understand the terms

    • @kevinwoodrobotics
      @kevinwoodrobotics Рік тому

      Keypoint is the feature or point of interest. Descriptor is a way you uniquely describe the keypoint, which in the case of SIFT it is described by a vector of the orientation histogram of neighboring points.

  • @ranati2000
    @ranati2000 2 роки тому

    error: (-5:Bad argument) in function 'drawKeypoints'
    > Overload resolution failed:
    > - Can't parse 'keypoints'. Sequence item with index 0 has a wrong type
    > - Can't parse 'keypoints'. Sequence item with index 0 has a wrong type
    Sir Iam having error on using drawKeyPoints
    My Code
    orb = cv2.ORB_create()
    kp1 = orb.compute(img1, None)
    kp2 = orb.compute(img2, None)
    imgKp1 = cv2.drawKeypoints(img1, kp1, None, flags=0)
    imgKp2 = cv2.drawKeypoints(img2, kp2, None, flags=0)

    • @DigitalSreeni
      @DigitalSreeni  2 роки тому +1

      Not sure of the error. One suggestion is to convert your images and keypoints to integer and see if that helps.

    • @ranati2000
      @ranati2000 2 роки тому

      @@DigitalSreeni ok thanks i shall do that

  • @jijie133
    @jijie133 2 роки тому

    Great video!

  • @adammarcaidajr.9478
    @adammarcaidajr.9478 4 роки тому

    thank you

  • @seaweedseaside5905
    @seaweedseaside5905 5 років тому

    Amazing! Thanks

  • @lorizoli
    @lorizoli 3 роки тому

    This is great content. I'm just impatient to go faster. But I cant!!!

  • @whomilv
    @whomilv 2 роки тому

    The patent of SIFT expired in the year 2020. It is available thru cv2.SIFT_create() function.

  • @edmondVerboseLogging
    @edmondVerboseLogging 3 роки тому

    thank you

  • @hamidmajidibalanji691
    @hamidmajidibalanji691 2 роки тому

    Thank you