Low-Quality Image Detection by Hierarchical VAE

TL;DR

This paper introduces an unsupervised method using hierarchical variational autoencoders to detect low-quality images with various degradations, providing visual clues to assist human recognition, outperforming existing out-of-distribution detection methods.

Contribution

The study presents a novel unsupervised approach leveraging hierarchical VAEs for low-quality image detection and visual explanation, addressing a new task in image quality assessment.

Findings

Outperforms several existing unsupervised out-of-distribution detection methods

Provides visual clues that help humans recognize low-quality images

Effective across various types of image degradation

Abstract

To make an employee roster, photo album, or training dataset of generative models, one needs to collect high-quality images while dismissing low-quality ones. This study addresses a new task of unsupervised detection of low-quality images. We propose a method that not only detects low-quality images with various types of degradation but also provides visual clues of them based on an observation that partial reconstruction by hierarchical variational autoencoders fails for low-quality images. The experiments show that our method outperforms several unsupervised out-of-distribution detection methods and also gives visual clues for low-quality images that help humans recognize them even in thumbnail view.