Prompt To Gesture: Synthetic and Real Videos for Deictic Gesture Recognition

Prompt To Gesture is a computer vision dataset built for training models to recognize deictic (pointing) gestures in human-robot interaction. It mixes real human recordings with image-to-video generated synthetic videos to address the scale and diversity limitations of traditional data collection. The dataset was created by author sano90 and last updated on Hugging Face in May 2026.

Use Cases

• Train deictic gesture recognition models based on the mixed real and synthetic video data.
• Evaluate model robustness in HRI settings based on the dataset's described focus.
• Benchmark synthetic data generation techniques for computer vision tasks.
• Develop human-robot interaction systems that rely on pointing gesture understanding.

Strengths

• Addresses a known bottleneck in gesture data collection by mixing real and synthetic sources.
• Explicitly built for a specific, applied task: deictic gesture recognition in human-robot interaction.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license are unknown, which limits suitability assessment.
• Data may reflect bias inherent to the specific real and synthetic sources used.

Provenance

Source

huggingface user sano90

Collection Method

Likely contains a mix of recorded human sessions and image-to-video generated synthetic videos.

Freshness

Last updated 2026-05-22 12:02:46.