Skip to content

This repository contains code used to perform image retrieval using transformers. It is a demonstration of how using vision transformers, metric learning, and a novel loss based on differential cross-entropy can lead to better retrieval than classical CNN-based methods.

Notifications You must be signed in to change notification settings

danilodjor/image-retrieval-using-transformers

Repository files navigation

MSc Thesis: "Towards improving the deep networks for image retrieval"

Github repository containing the code for the master thesis titled "Towards Improving the Deep Networks for Image Retrieval"

About

Image retrieval is a computer vision task of querying a database of images with an image and returning K most similar images from the database, according to some metric. K can be an arbitrary number of images.

The goal of this thesis is to explore the possiblities, the benefits, as well as potential downsides to using the transformer architecture to solve the problem of image retrieval.

About

This repository contains code used to perform image retrieval using transformers. It is a demonstration of how using vision transformers, metric learning, and a novel loss based on differential cross-entropy can lead to better retrieval than classical CNN-based methods.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published