-2

I'm following some online courses on artificial intelligence. I have a project in mind.

There will be a thousand pictures of people or objects in two folders. But these objects are unrelated things.

First folder name a and second folder name b.

Later, when I send an image, I want to know if it looks more like the images in folder a or b.

I created a model in Hugginface but they didn't mention what algorithm they used there.

Problem type: Binary Classification Model ID: 2043767063 CO2 Emissions (in grams): 1.6736

xcodertanks
  • 1
  • 3

1 Answers1

0

SIFT descriptors are quick compared to training on a CNN. So for simple, efficiently, and robust solutions to image comparison they would be ideal. In this case, instead of training a CNN on a set of images and using the penultimate layer’s output as the image representation; You could simply compute SIFT descriptors for all the images, and encode each image’s SIFT descriptors as a Fisher vector or VLAD.

SIFT (and related interest point descriptors such as ORB, SURF, and BRIEF), they provide a relatively simply, efficient solution.

Carter Pickett
  • 21
  • 5