RoadTones: 51k Tone-Aware Captions for Road Event Videos

RoadTones-51k is a multimodal dataset featuring 51,000 tone-aware captions for road event videos. It was created by siddhi-lipare and accompanies the paper 'RoadTones: Tone Controllable Text Generation from Road Event Videos'. The dataset was last updated on May 23, 2026.

Use Cases

• Train tone-controlled video captioning models based on the 51k tone-aware captions.
• Analyze the relationship between video content and tonal annotations across 215 personality traits.
• Develop specialized captioning systems for stakeholders like transport authorities based on the 16 writing styles.
• Benchmark multimodal models on generating contextually appropriate text for road safety scenarios.

Strengths

• Contains 51,000 annotated video-caption pairs.
• Provides rich tonal annotations across 215 distinct personality traits and 16 writing styles.
• Addresses a specific limitation in existing video-language models by enabling tone control.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface dataset by author siddhi-lipare.

Collection Method

Created to accompany the academic paper 'RoadTones: Tone Controllable Text Generation from Road Event Videos'.

Freshness

Last updated 2026-05-23 07:08:40; freshness should be verified.