Training data collection system for a learning-based photographic aesthetic quality inference engine

  • Authors:

  • Razvan Orendovici;James Z. Wang

  • Affiliations:

  • The Pennsylvania State University, University Park, PA, USA;The Pennsylvania State University, University Park, PA, USA

  • Venue:
  • Proceedings of the international conference on Multimedia
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

We present a novel data collection system deployed for the ACQUINE - Aesthetic Quality Inference Engine. The goal of the system is to collect online user opinions, both structured and unstructured, for training future generation learning-based aesthetic quality inference engines. The development of the system was based on an analysis of over 60,000 user comments of photographs. For photos processed and rated by our engine, all users are invited to provide manual ratings. The users can also choose up to three key photographic features that the user liked, from a list, or to add features not in the list. Within a few months that the system is available for public used more than 20,000 photos have received manual ratings and key features for over 1,800 photos have been identified. We expect the data generated over time will be critical in the study of computational inferencing of visual aesthetics in photographs. The system is demonstrated at http://acquine.alipr.com