TimeLens-100K: A Large-Scale Video Temporal Grounding Dataset

TimeLens-100K is a large-scale training dataset for video temporal grounding, created by TencentARC. The dataset was proposed in the paper 'TimeLens: Rethinking Video Temporal Grounding with Multimodal LLMs' and annotated using an automated pipeline powered by Gemini-2.5-Pro. It was last updated on December 19, 2025.

Use Cases

• Train video temporal grounding models based on the dataset's temporal annotations.
• Benchmark multimodal LLMs on video understanding tasks using the described high-quality data.
• Develop automated video annotation pipelines leveraging the Gemini-2.5-Pro methodology described.

Strengths

• Described as 'large-scale, diverse, and high-quality' for its intended task.
• Annotations were generated using an automated pipeline powered by Gemini-2.5-Pro.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and dataset size are unknown, which may limit suitability assessment.
• License information is unknown, which restricts clarity on permissible use.

Provenance

Source

TencentARC

Collection Method

Annotations were created using an automated pipeline powered by Gemini-2.5-Pro.

Freshness

Last updated 2025-12-19 15:05:09.