Datasets:
| language: | |
| - zh | |
| task_categories: | |
| - translation | |
| - video-classification | |
| pretty_name: CSL-Daily Video and Text Annotations | |
| # CSL-Daily | |
| This repository packages CSL-Daily videos with train, dev, and test metadata for use with Hugging Face `VideoFolder`. | |
| Each split contains a `metadata.parquet` file. Its `file_name` column is the relative path to the corresponding MP4. Video files are grouped by the first four characters of `clip_id` so that every directory remains below the Hugging Face per-folder file limit. | |
| ## Metadata columns | |
| - `file_name`: relative path to the MP4 within the split | |
| - `clip_id`: CSL-Daily video identifier | |
| - `translation`: Chinese sentence translation | |
| - `gloss`: gloss sequence | |
| - `language`: language name | |
| - `video_num_frames`: number of encoded video frames | |
| - `video_fps`: video frame rate | |
| - `video_duration_s`: video duration in seconds | |
| The record for `S000005_P0004_T00` was absent from the available derived annotation sources. Its translation and gloss were inferred from the two other recordings with the same sentence identifier, `S000005`. | |
| ## Access and licensing | |
| CSL-Daily is distributed under its original provider's access agreement. Users must obtain authorization from the CSL-Daily owners and comply with the original terms. Keep this repository private unless redistribution permission has been granted. | |