VISTA-400K: Synthetic Video Instruction-Following Data for LMMs

TIGER-Lab's VISTA-400K is a synthetic dataset designed to enhance video understanding in large multimodal models. The data is generated by the VISTA pipeline, which creates long-duration, high-resolution video instruction-following examples. The dataset was last updated on the Hugging Face platform in December 2024.

Use Cases

• Training video-language models based on the described synthetic instruction-following data.
• Benchmarking model performance on video understanding tasks based on the generated high-resolution, long-duration examples.
• Augmenting existing video datasets with synthetic instruction-following pairs based on the VISTA pipeline methodology.

Strengths

• Data is generated by a dedicated pipeline (VISTA) for creating instruction-following examples.
• Dataset focuses on long-duration and high-resolution video content as described.
• Authored by TIGER-Lab, a known research entity in the field.

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 license information are unknown, which may limit suitability assessment.

Provenance

Source

TIGER-Lab via Hugging Face.

Collection Method

Synthesized via the VISTA video spatiotemporal augmentation pipeline.

Time Range

null

Freshness

Last updated 2024-12-19 22:33:04.

Geography

null