SA-Co/VEval: Multimodal Video Evaluation Dataset from Three Domains

SA-Co/VEval is an evaluation dataset comprising three distinct video domains: SA-V, YT-Temporal-1B, and SmartGlasses. The dataset, created by Facebook and last updated in November 2025, includes validation and test splits for each domain.

Use Cases

• Benchmarking video understanding models based on the three distinct evaluation domains
• Evaluating temporal analysis algorithms using the YT-Temporal-1B video subset
• Testing models on egocentric video data from the SmartGlasses subset
• Comparing model performance across different video sources and splits

Strengths

• Dataset is structured with separate validation and test splits for each of its three domains
• Includes videos from three distinct sources: SA-V, YT-Temporal-1B, and SmartGlasses
• Last updated on 2025-11-17, indicating recent maintenance

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download
• Column-level documentation is absent; field semantics must be inferred after download
• Row count, file formats, and specific column details are unknown

Provenance

Source

Facebook

Collection Method

Videos are sourced from the SA-V dataset, YT-Temporal-1B, and SmartGlasses recordings.

Time Range

null

Freshness

Last updated 2025-11-17 17:53:51

Geography

null