The problem RISE - Reverse Image Search Engine solves

It is a pretty famous way of finding similar images, images which are equivalent to the given input image. The image will be queried in the database and a set of images related to the input image will be shown. The method of Reverse Image Search is being used by large Tech Giants such as Google all across the world.
This solution can be used by various industries as well as small scale sellers. They can use this technique to search for a particular product just by its image, rather than generating unique IDs for each and every product and storing them on the server. In Large Scale Industries, the inventory can be managed by this solution in a very efficient manner. Similarly in local shops and vendor stalls, they can setup a system to watch over the amount of items left in the store (Say for example a fruit seller can watch the quantity of apples remaining in the Store by just clicking a photo of apples and uploading it).

Challenges we ran into

The System design generated for this product was quite a large issue. We had to first connect a dataset where we can search the images. This was done using an S3 bucket (AWS) and connecting it to our backend AI models.
The AI models used here are ViT-GPT-2, which is used to generate captions for the images that are being sent to the server. It transforms the Images into a suitable caption and then generates the final description of the image. Setting up this model was a big task, and monitoring it for its efficiency and accuracy was a major drawback in the start. However, we overcame these issues during the first day of the hackathon.
The next task was to connect all the different parts through a single backend to make it into a single application, which was a cumbersome issue. We used Flask for this task, as it helped us integrate all these into a single application and without any major issues.