Fig 1.
Paintings from the MultitaskPainting100k dataset.
Each row shows samples from a different artist. We included different styles and genres based on the color coding. We state that all the paintings are all in the public domain(courtesy wikiart.org).
Table 1.
State-of-the-art results for artist, style and genre categorization, including samples and classes in each tasks.
Fig 2.
Comparing the block of ResNet, Res2Net, ResNeXt and ResNest.
A. ResNet block. B. Res2Net block. C. ResNeXt block. D. ResNest block.
Fig 3.
The overall general architecture of the model.
We modified the last full connection layer of CNN to meet the different categories of paintings in different tasks. There, we show the style categories. We state that all the paintings are all in the public domain(courtesy wikiart.org).
Table 2.
The results of the classification in Painting-91.
Table 3.
The results of the classification in Wikiart.
Table 4.
The results of the classification in MultitaskPainting100k.
Fig 4.
Embedding of the paintings projected in Painting-91 and WikiArt using t-SNE.
Each node is a painting, and the coloring is mapped to the style attribute and genre attribute. A. Painting-91 with style attribute. B. WikiArt with genre attribute.
Fig 5.
Confusion matrix of the different classification tasks.
A. style classifications in Painting-91 using Resnet50. B. style classifications in Painting-91 using EfficientNet. C. genre classifications in WikiArt using Resnet50. D. genre classifications in WikiArt using EfficientNet.
Fig 6.
Similarity search for the painting Starry Night, by Vincent Van Gogh.
A. The similarity results show the top six paintings retrieved using artist features. B. The similarity results show the top six paintings retrieved using style features. We state that all the paintings are all in the public domain(courtesy wikiart.org).