1
$\begingroup$

I have a set of images of various products from different websites. I want to cluster the images based on the product shown in the image. How can I generate a suitable feature vector for an image for this purpose?? I just need to know how to generate a feature vector given an image. I tried NetVLAD, but it is very slow. I would like something that is fast and gives high accuracy for clustering in the scenario I have described. Please help me.

$\endgroup$
  • $\begingroup$ You should add more info about your dataset (size, number of samples, number of desired clusters). You should also explain what do you mean by slow/fast and high accuracy. These things mainly depend on your computational limitations... $\endgroup$ – Mark.F Jan 8 '19 at 11:42
  • $\begingroup$ There are 282,082 images with a total size of almost 60 GB. I don't have a desired number of clusters. I want an accuracy of better than 80%. I want it to be fast enough to run on a workstation in a couple of days (at most). $\endgroup$ – user65516 Jan 9 '19 at 15:06
  • $\begingroup$ Can someone help me with this?? $\endgroup$ – user65516 Jan 10 '19 at 6:25
0
$\begingroup$

Feature extraction is basically reducing the amount of resources required to describe a large set of data. Analysis with a large number of variables (like images) generally requires a big amounts of memory and computation power. You can start from the simplest and slowly work your way up (in computational requirements) until you reach your desired accuracy.

Start from:

ORB (oriented FAST, rotated robust independent features) -> SURF (speed up robust feature) -> SIFT (scale invariant feature transform) -> OverFeat (Convolutional Network) -> vgg16 ... (There are many more, any neural network model can be used for this)

$\endgroup$
  • $\begingroup$ I would like to try out ORB. But, I am facing a difficulty. Given two different images, how do I find the distance between them using ORB features?? $\endgroup$ – user65516 Jan 16 '19 at 15:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.