Home // CONTENT 2023, The Fifteenth International Conference on Creative Content Technologies // View article

Using Text Queries to Look Up Unlabeled Images: A Command-Line Search Tool Based on CLIP

Authors:
Yurij Mikhalevich

Keywords: image search, image indexing, photo management, computer vision, natural language processing

Abstract:
This paper presents a practical, scalable implementation of an image search engine using OpenAI’s Contrastive Language-Image Pre-Training (CLIP) model. The method pro- vides a convenient Command-Line Interface (CLI) and introduces a cache layer powered by SQLite 3 relational database management system (RDBMS) that facilitates efficient repetitive image searches within extensive image databases using natural language queries. The method’s effectiveness was evaluated on ImageNet-1k and CIFAR-100 datasets, yielding a 31.17% top-1 accuracy on the ImageNet-1k train set and 55.15% top-1 accuracy on the CIFAR-100 test set. The scalability study showed that indexing time scales linearly with the number of images, and image search time increases only slightly; for example, on an Apple M1 Max CPU, indexing over a million images took 26.36 times more than indexing 50,000 images, while querying the larger image set took just 2.75 times longer. This approach is particularly relevant for industries managing vast volumes of visual data, such as media and entertainment, security, and healthcare.

Pages: 7 to 11

Copyright: Copyright (c) IARIA, 2023

Publication date: June 26, 2023

Published in: conference

ISSN: 2308-4162

ISBN: 978-1-68558-048-3

Location: Nice, France

Dates: from June 26, 2023 to June 30, 2023