Behaviour of Recent Aesthetics Assessment Models with Professional Photography

Mathieu Chambe (1) , Rémi Cozot (2) , Olivier Le Meur (1)

1 PERCEPT - Computational Visual Perception and Applications
2 LISIC - Laboratoire d'Informatique Signal et Image de la Côte d'Opale

Aesthetic quality assessment for photographs is an important research topic since it can be used by a number of applications, such as image database management or image browsing. In 2012, the Aesthetic Visual Analysis (AVA) dataset has been proposed. Those 255,000 aesthetically annotated images are a key ingredient for training and for testing new models for aesthetics prediction. As AVA dataset is mainly composed of competitive photographs, we evaluate whether or not those computational models of aesthetics generalize well and perform well over professional photographs. We notice that the different models we test behave quite differently. Besides, we fine-tune the model using professional photographs and the results show that this process is effective.

Mots clés: Aesthetics Assessment; Deep Learning; Over- fitting; Professional Photography

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]; Traitement des images [eess.IV]; Réseau de neurones [cs.NE]

Mathieu Chambe, Rémi Cozot, Olivier Le Meur. Behaviour of Recent Aesthetics Assessment Models with Professional Photography. 2019. ⟨hal-02374494⟩