VideoTemp Bench: Video Question Answering with Agentic Temporal Grounding

A benchmark for video question answering and temporal grounding, likely sourced from the NExT-GQA dataset. The dataset was created by Kwai-Keye and last updated on Hugging Face on 2026-05-20. It is designed to evaluate models that perform on-demand temporal grounding to locate relevant video segments before answering questions.

Use Cases

• Benchmarking video question-answering models based on the described agentic pipeline.
• Evaluating temporal grounding algorithms based on the iterative refinement process mentioned.
• Training models for multimodal reasoning that integrate visual evidence with natural language queries.

Strengths

• Dataset is associated with a published research concept (VideoTemp-o3) for harmonizing temporal grounding and video understanding.
• Last update timestamp is explicitly provided: 2026-05-20 12:40:21.

Limitations

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

Provenance

Source

NExT-GQA (https://github.com/doc-doc/NExT-GQA)

Freshness

Last updated 2026-05-20 12:40:21; freshness should be verified.