Doundobench: Egocentric Video Action Annotations with Forward and Reverse Prompts

Action annotations for egocentric video clips include original narrations, temporal boundaries, verb and noun labels. The dataset pairs natural-language prompts describing a forward action and its reverse or undo action. It was created by 'doundo' and last updated on Hugging Face in May 2026.

Use Cases

• Train models for action recognition in egocentric video based on verb and noun labels.
• Develop AI systems that understand and generate descriptions of reversible actions based on paired forward/undo prompts.
• Benchmark temporal action localization models using the provided temporal boundaries.
• Research the relationship between natural language narration and visual action sequences.

Strengths

• Annotations include paired natural-language prompts for forward and reverse actions, enabling bidirectional reasoning tasks.
• Each example contains multiple annotation types: original narration, temporal boundaries, and verb/noun labels.
• Last updated on 2026-05-06, suggesting recent maintenance.

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

Freshness

Last updated 2026-05-06 06:18:03; freshness should be verified.