Wearable AI: Egocentric Video Question-Answering Benchmark from ECCV 2026

A benchmark of egocentric videos paired with three complementary video question-answering tasks for evaluating wearable-AI assistants. The dataset was released by Facebook as part of the Wearable AI Workshop at ECCV 2026. It includes baseline code and evaluation scripts for assessing models on real-world everyday activity videos.

Use Cases

• Benchmarking wearable-AI assistants based on the described video question-answering tasks.
• Training models for egocentric video understanding based on first-person activity videos.
• Developing multi-task learning systems based on the three complementary evaluation tasks mentioned.

Strengths

• Dataset is associated with a peer-reviewed workshop at a major computer vision conference (ECCV 2026).
• Includes baseline code and evaluation scripts, which suggests a standardized benchmark setup.

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

facebook

Collection Method

Likely collected from head-mounted wearable cameras capturing real-world everyday activities.

Freshness

Last updated 2026-05-21 21:57:28; freshness should be verified.