Classificação de pinturas com aprendizado de máquina

• Autor(a) principal: Vinhas, Lucas Gianoglio
• Data de Publicação: 2021
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/213838
• topic: Machine learning; Convolutional Neural Network; Image recognition; Classification; Aprendizado de máquina; Reconhecimento imagens; Classificação

This work intends to explore different architectures of Convolutional Neural Networks to classify paintings according to their authors. Such is the complexity of the problem, it was necessary to use the technique in which consists of exhaustively training the network with millions of samples on a given basis and then transfer this knowledge to the proposed problem. We made use of five network architectures: VGG16, Inception, Xception, ResNet and InceptionResnet. The dataset used was the WikiArt Dataset, in which all painters with more than 19.000 works were selected, totaling 23 classes for categorization. were balanced according to the number of images present. It was evident in the experiments the effectiveness of Inception network and its variants. All training was carried out on Google’s Colab platform, in order to optimize training time, which is not short on home computers.