OmniArt

Overview

Baselines are the starting point of any quantitative multimedia research, and benchmarks are essential for pushing those baselines further. OmniArt is a large-scale artistic benchmark dataset featuring over 2 million images with rich structured metadata, aggregated from multiple museum collections around the world.

OmniArt contains annotations for dozens of attribute types and features semantic context information through concepts, IconClass labels, color information, and object-level bounding boxes. The dataset establishes baseline scores on multiple tasks such as artist attribution, creation period estimation, type, style, and school prediction.

In addition to metadata-related experiments, we explore the color spaces of art through different types and evaluate a transfer learning object recognition pipeline. The dataset was designed for seamless integration with popular deep learning frameworks including PyTorch, Keras, and TensorFlow.

Dataset Statistics

Metadata Types

Featured Collections

OmniArt aggregates artwork from some of the world's most prestigious institutions:

Benchmark Tasks

The dataset includes seven specialized subsets for evaluating different aspects of artistic understanding:

ArtSight Exploration Tool

Framework Support

OmniArt provides custom data loaders for seamless integration with popular deep learning frameworks:

Publications

Citation

@article{strezoski2018omniart,
  title     = {OmniArt: A Large-scale Artistic Benchmark},
  author    = {Strezoski, Gjorgji and Worring, Marcel},
  journal   = {ACM Trans. Multimedia Comput. Commun. Appl.},
  volume    = {14},
  number    = {4},
  articleno = {88},
  pages     = {1--21},
  year      = {2018},
  publisher = {ACM},
  doi       = {10.1145/3273022}
}