MovieChat is a dataset for long video understanding, introduced in a CVPR 2024 research paper. The dataset is associated with a method that converts dense video tokens into sparse memory for efficient processing. The repository was last updated on 2025-01-29 by author wenhaochai under a BSD-3-Clause license.
Use Cases
- Training models for long video question answering based on the described sparse memory framework.
- Benchmarking video understanding architectures on tasks requiring temporal reasoning.
- Developing efficient tokenization methods for video data as suggested by the dense-to-sparse conversion approach.
Strengths
- Associated with a peer-reviewed CVPR 2024 publication, indicating academic rigor.
- Repository was updated on 2025-01-29, suggesting recent maintenance.
- Released under the permissive BSD-3-Clause license.
Limitations
- Description metadata is limited; actual data quality, size, and format require manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count and dataset scale are unknown, which may limit suitability assessment.
Provenance
- Source
- github repository by wenhaochai.
- Collection Method
- Likely created for the CVPR 2024 research paper 'MovieChat: From Dense Token to Sparse Memory for Long Video Understanding'.
- Time Range
- null
- Freshness
- Last updated 2025-01-29 11:16:02; freshness should be verified.
- Geography
- null