How does rubrix work?

rubrix makes use of multiple deep learning models to do several complicated image/text related tasks.

1. Image Search

• Architecture

• Video Summary

Click on the GIF below to learn how image search in rubrix works:

2. Reverse Image Search

• Architecture

• Summary

  The reverse image search algorithm in rubrix can be broken down into three main steps:

  • Using an image recognition model, extract feature vectors for all the images in the database.
  • When the user uploads an image:
    1. Extract all the objects in the image using YOLOv4 algorithm.
    2. From the inverse image index mapping different objects to the images containing them, prune to extract relevant images, i.e. those containing the objects as in the user-uploaded image.
  • Calculate the similarity scores of the feature vector corresponding to the user-uploaded image with those corresponding to all pruned images, and return the top-5 results.