Flickr30k

The 'Flickr 30K' dataset contains 30,000 images, each paired with five descriptive captions, and is used to evaluate image captioning models by assessing their ability to generate coherent and semantically meaningful textual descriptions for given images.

Papers using Flickr30k (3)

Do All Visual Tokens Matter Equally? Object-Evidence Preserving Token Merging for Vision-Language Retrieval2026AGIC: Attention-Guided Image Captioning to Improve Caption Relevance2025Position Focused Attention Network For Image-text Matching2019 · 198 cites