COYO-700M: 747 Million Image-Text Pairs for Vision-Language Training

COYO-700M is a large-scale dataset containing 747 million image-text pairs with additional meta-attributes. It was created by KakaoBrain using a strategy of collecting informative alt-text and associated images from HTML documents. The dataset was last updated on August 30, 2022.

Use Cases

• Training image captioning models based on the large collection of image-text pairs.
• Training text-to-image generation models based on the alt-text and image associations.
• Performing zero-shot classification tasks based on the multimodal features.
• Training or fine-tuning multimodal models based on the dataset's scale and meta-attributes.

Strengths

• Contains 747 million image-text pairs, indicating a large scale.
• Includes many other meta-attributes to increase usability, as mentioned in the description.
• Follows a collection strategy similar to established vision-and-language datasets.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Last updated 2022-08-30 19:07:52; freshness should be verified.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

KakaoBrain

Collection Method

Collected from informative pairs of alt-text and associated images in HTML documents.

Time Range

null

Freshness

2022-08-30 19:07:52

Geography

null