SpatialVID: A Large-Scale Video Dataset with Spatial Annotations

SpatialVID is a large-scale video dataset featuring spatial annotations, as described in a CVPR 2026 paper. The dataset was created by researchers from Nanjing University and the Institute of Automation, Chinese Academy of Science. Its specific scale, such as the number of videos or annotations, is not detailed in the provided metadata.

Use Cases

• Training object detection models based on the spatial annotations mentioned in the description.
• Developing video understanding algorithms for scene geometry based on the dataset's spatial features.
• Benchmarking performance of models on tasks requiring spatial reasoning in video data.

Strengths

• Associated with a CVPR 2026 research paper, indicating a peer-reviewed academic foundation.
• Created by a collaboration between Nanjing University and the Chinese Academy of Sciences, suggesting institutional backing.

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 and dataset size are unknown, which may limit suitability assessment.

Provenance

Source

Hugging Face user FelixYuan, associated with academic authors from Nanjing University and the Institute of Automation, Chinese Academy of Science.

Freshness

Last updated 2026-03-24 08:55:31; freshness should be verified.