hungpham2002 / image-retrieval-using-pretrained-vit

Apply pre-trained Vision transformer to retrieve images

image-retrieval-using-pretrained-ViT

Apply pre-trained Vision transformer to retrieve images

Introduction:

An image retrieval system is a computer system used for browsing, searching and retrieving images from a large database of digital images.

Project Statement:

Given a list of images, build an image retrieval system that finds relevant or similar images among the source images compared to the query image.

Modern Aproaching:

Feature Extractor: A component (typically a network) that is automatically identifying and extracting meaningful features from raw image data. Some well-known extractor are:

• Digital Image Processing: HOG, SIFT...
• Traditional Machine Learning: Bag of Visual Words, PCA...
• Deep Learning: CNNs, Autoencoder, ViT...

Apply pre-trained ViT:

Vision Transformer (ViT): A deep learning architecture designed for image classification tasks. ViT deviates from traditional CNNs by using a transformer-based architecture.

❖ Paper: https://arxiv.org/pdf/2010.11929.pdf

Idea

Dataset:

Download dataset or add shortcut if using Google Colab: here

Demo