Summary

IEICE Information and Communication Technology Forum

2020

Session Number:ICTF_5

Session:

Number:ICTF2020_paper_24

Invertible local image descriptors learned with variational autoencoders

Nina Zikakic,  Aleksandra Pizurica,  

pp.-

Publication Date:2021/03/24

Online ISSN:2188-5079

DOI:10.34385/proc.64.ICTF2020_paper_24

PDF download (1.1MB)

Summary:
In this paper, we propose an efficient method for learning local image descriptor and its inversion function using a variational autoencoder (VAE). We design a loss function of the VAE specifically for this purpose, which, on one hand, incentivises the similarities between input patches to be preserved in latent space, and on the other hand, ensures good reconstruction of the patches from their encodings in latent space. Our proposed descriptor demonstrates better patch retrieval compared to the reference autoencoder-based local image descriptor, and also shows improved reconstruction of patches from their encodings.