VidLLVIP: Temporally and Spatially Aligned Infrared-Visible Video Pairs

VidLLVIP is an unofficial processed dataset derived from the raw LLVIP videos. The dataset provides temporally aligned, spatially registered, and quality-checked 5-second video pairs. It was created by user jianfeng0369 and last updated on Hugging Face in May 2026.

Use Cases

• Training video fusion models based on aligned infrared-visible pairs.
• Developing cross-modal registration algorithms based on spatially registered frames.
• Advancing multimodal video understanding based on temporally synchronized sequences.

Strengths

• Video pairs are temporally aligned and spatially registered.
• Data has undergone a quality-checking process.
• Each video clip is 5 seconds long, providing a consistent unit for analysis.

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 is unknown, which may limit suitability assessment.

Provenance

Source

Derived from the raw LLVIP videos.

Collection Method

Processed to provide temporal alignment, spatial registration, and quality checking.

Time Range

null

Freshness

Last updated 2026-05-06 04:11:46; freshness should be verified.

Geography

null