Universal Image Embedding: Retaining and Expanding Knowledge With Multi-Domain Fine-Tuning

Universal Image Embedding: Retaining and Expanding Knowledge With Multi-Domain Fine-Tuning

The overall purpose of this study is to propose a novel fine-tuning method for the CLIP architecture that enables the retention of pre-existing knowledge from large datasets and the creation of a domain-agnostic image encoder for universal image embedding, addressing the challenge of transferring kn...

Full description

Bibliographic Details
Main Authors: Socratis Gkelios, Anestis Kastellos, Yiannis S. Boutalis, Savvas A. Chatzichristofis
Format: Article
Language:English
Published: IEEE 2023-01-01
Series:IEEE Access
Subjects:
CBIR
vision transformers
CLIP
deep learning
global features
image retrieval
Online Access:https://ieeexplore.ieee.org/document/10103696/

Internet

https://ieeexplore.ieee.org/document/10103696/

Similar Items