Act-Env Matching: Activity and Context-Aware Avatar Positioning for AR Telepresence

A research dataset accompanies a paper proposing Act-Env Matching, a method for positioning life-sized avatars in augmented reality telepresence. The method uses a multimodal large language model to determine avatar placement based on a remote user's activity label and a video of the local environment. The dataset, authored by Hideki Deguchi and last updated in May 2026, likely contains experimental results and demonstration materials supporting the proposed system.

Use Cases

• Training activity-aware avatar placement models based on the described Act-Env Matching method.
• Benchmarking AR telepresence systems against the proposed context-aware positioning approach.
• Analyzing user perception of avatar plausibility in heterogeneous environments as described in the experimental results.

Strengths

• The dataset supports a novel method designed for heterogeneous environments, removing the requirement for similar spatial layouts.
• The accompanying research includes results from two experiments and a real-environment demonstration.

Limitations

• The dataset description is general; column-level documentation and sample data are unavailable.
• The row count is unknown, which may limit suitability assessment for large-scale model training.

Provenance

Source

Hideki Deguchi

Collection Method

Likely derived from experiments and a demonstration described in the associated research paper.

Freshness

Last updated 2026-05-12 04:19:36