NautData: 1.45 Million Underwater Image-Text Pairs for Instruction Tuning

NautData is a large-scale underwater instruction-following dataset containing 1.45 million image-text pairs. It was constructed by H-EmbodVis to bridge the gap in large-scale underwater multi-task instruction-tuning datasets. The dataset was introduced in the paper NAUTILUS and is intended for advancing underwater scene understanding methods.

Use Cases

• Training underwater vision-language models based on the described image-text pairs.
• Benchmarking multimodal model performance on underwater scene understanding tasks.
• Instruction-tuning large models for specific underwater applications based on the dataset's multi-task focus.

Strengths

• Contains 1.45 million image-text pairs, indicating a substantial scale.
• Explicitly designed for multi-task instruction-tuning, a specific and advanced use case.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified; last updated 2025-12-18 15:02:44.

Provenance

Source

H-EmbodVis

Collection Method

Constructed for the NAUTILUS paper; specific collection method not detailed.

Freshness

2025-12-18 15:02:44