Transfer learning of ImageNet Object Classification Challenge features to image aesthetic binary classification

  • Bruno Tinen Escola Politécnica da USP
  • Jun Okamoto USP

Resumo


The aesthetic classification of photographies is a problem of separating aesthetically pleasing images from not pleasing images using algorithms that describe and evaluate both emotional and technical factors. Since the mass adoption of deep convolutional neural network (DCNN) models for image classification problems different DCNN architectures have been developed due to its overall better performance, pushing the boundaries of the state-of-the-art performance of the image classification further. This paper evaluates how architectures and features that were primarily developed for the ImageNet Object Classification Challenge perform when analyzed under the aesthetic scope. A high level transfer learning model composed of a DCNN layer and a top layer that behaves as a linear SVM is proposed and seven different DCNN architectures are trained using it. Scenarios with just transfer learning and with fine tuning are evaluated and a model using the ResNet-Inception V2 architecture is proposed, which achieves results better than current state-of-the-art for the experiment conditions used.

Palavras-chave: Machine Learning, Artificial Neural Networks, Deep Learning, Computer Vision

Referências

Barrett, C., Fontaine, P., and Tinelli, C. (2016). The Satisfiability Modulo Theories Library (SMT-LIB). www.SMT-LIB.org, acessado em 13/01/2018.

Barrett, C., Sebastiani, R., Seshia, S., and Tinelli, C. (2009). Satisfiability modulo theories. In Biere, A., Heule, M. J. H., van Maaren, H., and Walsh, T., editors, Handbook of Satisfiability, volume 185 of Frontiers in Artificial Intelligence and Applications, chapter 26, pages 825–885. IOS Press.

Ghallab, M., Nau, D., and Traverso, P. (2004). Automated Planning: theory and practice. Elsevier.

Gregory, P., Long, D., Fox, M., and Beck, J. C. (2012). Planning Modulo Theories : Extending the Planning Paradigm. International Conference on Automated Planning and Scheduling, pages 65–73.

McDermott, D., Ghallab, M., Howe, A., Knoblock, C., Ram, A., Veloso, M., Weld, D., and Wilkins, D. (1998). Pddl-the planning domain definition language.

Rich, E. (1985). Artificial intelligence and the humanities. Computers and the Humanities, 19(2):117–122.

Russell, S. J. and Norvig, P. (2003). Artificial intelligence: a modern approach. Elsevier.

Weld, D. S. (1999). Recent Advances in AI Planning. AI Magazine, 20(2):93.

NATambém foi demonstrado a eficiência do PMTPlan, mesmo sendo um planejador em fase experimental em comparação com outro planejador, o MetricFF, ao resolver um problema clássico de planejamento com recursos. Em trabalhos futuros planeja-se aprimorar o planejador PMTPlan, acrescentando a ele métricas em relação a quantidade e, principalmente, tempo. Além disso, pretende-se criar novos domı́nios para o novo formalismo, no qual um deles foca no estudo de caso para problema de transporte ferroviário.

Barrett, C., Fontaine, P., and Tinelli, C. (2016). The Satisfiability Modulo Theories Library (SMT-LIB). www.SMT-LIB.org, acessado em 13/01/2018.

Barrett, C., Sebastiani, R., Seshia, S., and Tinelli, C. (2009). Satisfiability modulo theories. In Biere, A., Heule, M. J. H., van Maaren, H., and Walsh, T., editors, Handbook of Satisfiability, volume 185 of Frontiers in Artificial Intelligence and Applications, chapter 26, pages 825–885. IOS Press.

Ghallab, M., Nau, D., and Traverso, P. (2004). Automated Planning: theory and practice. Elsevier.

Gregory, P., Long, D., Fox, M., and Beck, J. C. (2012). Planning Modulo Theories : Extending the Planning Paradigm. International Conference on Automated Planning and Scheduling, pages 65–73.

McDermott, D., Ghallab, M., Howe, A., Knoblock, C., Ram, A., Veloso, M., Weld, D., and Wilkins, D. (1998). Pddl-the planning domain definition language.

Rich, E. (1985). Artificial intelligence and the humanities. Computers and the Humanities, 19(2):117–122.

Russell, S. J. and Norvig, P. (2003). Artificial intelligence: a modern approach. Elsevier.

Weld, D. S. (1999). Recent Advances in AI Planning. AI Magazine, 20(2):93.
Publicado
15/10/2019
Como Citar

Selecione um Formato
TINEN, Bruno; OKAMOTO, Jun. Transfer learning of ImageNet Object Classification Challenge features to image aesthetic binary classification. In: ENCONTRO NACIONAL DE INTELIGÊNCIA ARTIFICIAL E COMPUTACIONAL (ENIAC), 16. , 2019, Salvador. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2019 . p. 130-141. DOI: https://doi.org/10.5753/eniac.2019.9278.