OpenGVLab provides annotations and videos for training the VideoChat-Flash model. The repository includes instructions for concatenating video files from subsets like 'longvid_subset/coin_grounding_10k_zip'. This dataset was last updated on June 24, 2025.
Use Cases
- Training hierarchical compression models for long-context video based on the described annotations.
- Fine-tuning video-language models using the provided video and annotation pairs.
- Benchmarking video grounding and event understanding tasks on subsets like 'ego4dhcap_eventunderstanding_2k'.
- Researching methods for efficient long-video processing and summarization.
Strengths
- Dataset is directly associated with a published research paper (Li et al., 2024).
- Includes specific instructions for data preparation, such as concatenating zip files.
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 are unknown, which may limit suitability assessment.
Provenance
- Source
- OpenGVLab
- Collection Method
- Created for training the VideoChat-Flash model; specific collection method is not detailed.
- Time Range
- null
- Freshness
- Last updated 2025-06-24 07:51:17; freshness should be verified.
- Geography
- null