UCS-Bench: Egocentric Video QA for Continual Spatial Intelligence

170+ hours of streaming egocentric visual observations and 7K+ timestamped questions form this benchmark. It is designed to evaluate user-centric continual spatial intelligence in dynamic environments from a first-person point of view. The dataset was created by cocowy1 and was last updated on the Hugging Face platform in June 2026.

Use Cases

• Benchmarking models for spatial question answering based on timestamped questions in streaming video.
• Evaluating long-term spatial memory retention based on the requirement to understand dynamic environments.
• Testing real-time location and movement grounding based on questions tied to a user's egocentric perspective.

Strengths

• Contains over 170 hours of egocentric video data.
• Includes more than 7,000 timestamped questions for evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset size in rows and specific file formats are unknown.
• Last updated 2026-06-13 06:31:03; freshness should be verified.

Provenance

Source

huggingface

Collection Method

Likely collected for research purposes to create a benchmark for spatial intelligence.

Freshness

Last updated 2026-06-13 06:31:03.