Home // CONTENT 2024, The Sixteenth International Conference on Creative Content Technologies // View article

Automatic Semantic Image Tagging at Scale: AI-Powered Command-Line Tool Based on CLIP

Authors:
Yurij Mikhalevich

Keywords: image tagging; image indexing; photo management; computer vision

Abstract:
This study introduces an efficient and scalable solution for image tagging through the utilization of OpenAI’s Contrastive Language-Image Pre-Training (CLIP) model. It features a user-friendly Command-Line Interface (CLI) and leverages a caching mechanism for image features introduced by the rclip tool, which is powered by the SQLite 3 Relational DataBase Management System (RDBMS). This enables effective tagging of image catalogs already indexed by rclip. The performance of this method was tested on the ObjectNet dataset, achieving 27.22% accuracy for the top-1 prediction and 50.42% for the top-5 predictions. A scalability analysis revealed that both the indexing and tagging processes increase linearly with the image count. For instance, processing 50,273 unindexed images on an Apple M1 Max CPU was 31.66 times longer than tagging 965 unindexed images, and handling 50,273 indexed images was 31.15 times longer than 965 indexed images. This strategy is particularly beneficial for sectors that handle large amounts of visual content and require external processing tools, such as the media and entertainment, security, and healthcare industries.

Pages: 7 to 13

Copyright: Copyright (c) IARIA, 2024

Publication date: April 14, 2024

Published in: conference

ISSN: 2308-4162

ISBN: 978-1-68558-159-6

Location: Venice, Italy

Dates: from April 14, 2024 to April 18, 2024