Negation-Aware Grounding Annotations for Flickr30k Images

CoVAND provides annotations for a negation-aware visual grounding dataset built upon the Flickr30k corpus. The dataset was created by author 2na-97 to support the ICLR 2026 paper on negation-aware vision-language models. It was last updated in April 2026.

Use Cases

• Train vision-language models to detect objects mentioned with negation in image captions.
• Benchmark model performance on negation-aware visual grounding tasks using the provided annotations.
• Analyze the relationship between structured reasoning and token merging techniques for handling negation.
• Evaluate model sensitivity to negated versus affirmative phrases in image descriptions.

Strengths

• Annotations are derived from the established Flickr30k image-captioning dataset.
• Dataset is associated with a peer-reviewed ICLR 2026 conference paper.

Limitations

• The dataset size, row count, and specific annotation format are not provided.
• Access requires separate download of the original Flickr30k images from an external source.

Provenance

Source

Annotations are derived from the Flickr30k Denotation Graph.

Collection Method

Created for research on negation-aware vision-language models, as described in the associated paper.

Time Range

null

Freshness

null

Geography

null