Article citationsMore>>

Boudiaf, M., Rony, J., Ziko, I.M., Granger, E., Pedersoli, M., Piantanida, P. and Ayed, I.B. (2020) Metric Learning: Cross-Entropy vs. Pairwise Losses.

has been cited by the following article:

• TITLE: Data Covariance Learning in Aesthetic Attributes Assessment

  AUTHORS: Zhihong Chen

  KEYWORDS: Aesthetic Assessment, Aesthetic Attributes Assessment, Data Uncertainty, Data Covariance, Maximum Likelihood, Convolutional Neural Network

  JOURNAL NAME: Journal of Applied Mathematics and Physics, Vol.8 No.12, December 17, 2020

  ABSTRACT: Image Aesthetic Assessment (IAA) is a widely considered problem given its usefulness in a wide range of applications such as the evaluation image capture pipelines, sharing and storage techniques media, but the intrinsic mechanism of aesthetic evaluation is seldom been explored due to its subjective nature and the lack of interpretability of deep neural networks. Noticing that the photographic style annotations of images (i.e. the score of aesthetic attributes) are more objective and interpretable compared with the Mean Opinion Scores (MOS) annotations for IAA, we evaluate the problem of Aesthetic Attributes Assessment (AAA) as to provide complementary information for IAA. We firstly introduce the learning of data covariance in the field of Aesthetic Attributes Assessment and propose a regression model that jointly learns from MOS as well as the score of all aesthetic attributes at the same time. We construct our method by extending the scheme of data uncertainty learning and propose data covariance learning. Our method achieves the state-of-the-art performance on AAA without architectural design that trains in a totally end-to-end manner and can be easily extended to existing IAA methods.