Fine-grained painting classification

A lot of progress has been made in the domain of image classification in the deep learning era, however, not so much for paintings. Even though paintings are images they are very different from photographs and classification of paintings requires in-depth domain knowledge compared to classifying an object. This makes the task of fine-grained classification of paintings even harder. In this thesis, we evaluate the classification of paintings into its various styles, genres, artists and formulate the problem of dating paintings as a classification problem. We experiment with the standard networks available as baselines and then improve the classification models via multi-task learning. We also propose a novel architectural addition to the VGG network to do fine-grained classification. Our models beat the existing state-of-the-art classifiers by a big margin.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2018

Author dc:creator

• Kedia, Manav

Contributors dc:contributor

• Lazebnik, Svetlana

• Painting
• Classification
• Styles
• Artist
• Multi-task
• Visual Geometry Group (VGG)

Statement dc:rights

• Copyright 2017 Manav Kedia

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/99388

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/99388