SynGallery-1024: Synthetic Images of Real Paintings for Instance Recognition

24,490 high-resolution synthetic images render 4,898 real paintings from the MET Open Access collection in a 3D gallery. Each artwork is captured from 5 camera viewpoints at 1024×1024 resolution, paired with source photos and museum metadata. The dataset, created by patryk-bartkowiak, was last updated on June 17, 2026.

Use Cases

• Train instance-level artwork recognition models based on synthetic images paired with real painting photos.
• Benchmark domain adaptation techniques for models trained on synthetic data applied to real-world art images.
• Develop and test computer vision models for object detection and classification in procedurally generated 3D environments.
• Research the utility of synthetic data for tasks in digital humanities and cultural heritage preservation.

Strengths

• Contains 24,490 high-resolution (1024×1024) synthetic RGB images.
• Based on 4,898 real paintings sourced from the MET Open Access collection.
• Each painting is rendered from 5 distinct camera viewpoints, providing multiple perspectives.
• Images are paired with source photos and museum metadata like title, artist, date, and medium.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect the temporal and artistic bias inherent to the MET Open Access collection source.

Provenance

Source

Paintings sourced from the Metropolitan Museum of Art (MET) Open Access collection.

Collection Method

Real paintings were procedurally rendered in a randomized 3D art-gallery scene and captured from multiple camera viewpoints.

Freshness

Last updated 2026-06-17 13:47:38; freshness should be verified.