| license: cc-by-4.0 | |
| task_categories: | |
| - video-classification | |
| - other | |
| tags: | |
| - opencs2 | |
| - encoding | |
| - torchcodec | |
| - video | |
| - audio | |
| configs: | |
| - config_name: media_variants | |
| data_files: | |
| - split: train | |
| path: data/media_variants.parquet | |
| default: true | |
| # OpenCS2 Encoding Test | |
| Tiny dataset for validating OpenCS2 round-video encoding choices over real | |
| Hugging Face URLs. The `video` and `audio` columns are feature structs. For | |
| muxed variants they intentionally point to the same MP4, so torchcodec can | |
| decode video and audio from one HTTP object. | |
| Rows: 6 | |
| Recommended smoke test: | |
| ```python | |
| from datasets import load_dataset, Video, Audio | |
| ds = load_dataset("blanchon/opencs2_encoding_test", split="train") | |
| row = ds.cast_column("video", Video(seek_mode="approximate"))[0] | |
| frames = row["video"].get_frames_played_in_range(1.2, 2.2) | |
| samples = row["audio"].get_samples_played_in_range(1.2, 2.2) | |
| ``` | |