CommonCatalog CC-BY-NC: High-Resolution Flickr Images with Synthetic Captions

100 million high-resolution Creative Commons images collected from Yahoo Flickr users in 2014. The dataset contains images of up to 4k resolution, making it one of the highest resolution captioned image collections available. It was created by common-canvas and last updated on HuggingFace in May 2024.

Use Cases

• Training high-resolution image-to-text models based on the 4k resolution images.
• Benchmarking synthetic caption generation models using the provided captions.
• Analyzing the composition and content of large-scale Creative Commons image collections.
• Fine-tuning computer vision models on a diverse set of user-generated photographs.

Strengths

• Contains approximately 100 million images, indicating a large scale.
• Images are of up to 4k resolution, providing high visual detail.
• Explicitly sourced from a known platform (Yahoo Flickr) with a collection date (2014).

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect temporal and platform bias inherent to its 2014 Flickr source.

Provenance

Source

Yahoo Flickr users.

Collection Method

Collected from users of Yahoo Flickr under Creative Commons licenses.

Time Range

Collected in 2014.

Freshness

Last updated 2024-05-16 19:44:00; freshness should be verified.