tk_batch1

# TikTok Video Dataset (Batch 1) This dataset contains TikTok videos packed as tar shards for efficient bulk download. ## Overview | Item | Value | |------|-------| | Repository | `smile2333/tk_batch1` | | Tar shards | 3542 | | Leaf folders | 19 | | Videos | 1,766,243 | | Dominant resolution | 1080x1920 (98.4%) | | Duration median | 82.2s | | Pack format | `tiktok/video*/video_data_*/0000.tar` | Each tar usually contains 500 mp4 files. The last subpart in a leaf folder may contain fewer videos. ## Resolution | Resolution | Count | Ratio | |------------|------:|------:| | `1080x1920` | 1,737,325 | 98.4% | | `576x1024` | 24,812 | 1.4% | | `720x1280` | 1,314 | 0.1% | | `1920x1080` | 911 | 0.1% | | `3840x2160` | 245 | 0.0% | Orientation: landscape 1,157 (0.1%), portrait 1,763,780 (99.9%), square 0 (0.0%). ## Directory Layout ```text smile2333/tk_batch1/ README.md metadata.csv preview/ metadata.csv sample_001.mp4 ... manifest/ tar_index.csv tar_summary.json tiktok/ preview/ stats.json subparts.json meta/video2-28975H__video_data_02_1.json video2-28975H/ video_data_02_1/ 0000.tar 0001.tar ... video3-28645H/ video_data_03_1/ ... ``` Video files inside each tar are named by TikTok video id, for example `6984566817188564230.mp4`. Per-video metadata (`id`, resolution, duration, author, tar path) is in `tiktok/preview/meta/*.json`. ## Preview The `preview/` folder contains sample mp4 files and `metadata.csv`. ModelScope uses `metadata.csv` to render the dataset preview page. ## Download One Tar ```bash modelscope download smile2333/tk_batch1 tiktok/video2-28975H/video_data_02_1/0000.tar --repo-type dataset ``` ## Extract One Tar ```bash tar -xf 0000.tar ``` ## Load Sample Videos With MsDataset ```python from modelscope.msdatasets import MsDataset ds = MsDataset.load("smile2333/tk_batch1", subset_name="default", split="train") print(ds[0]) ``` ## Notes - Full dataset files are tar shards under `tiktok/video*/video_data_*` folders. - `manifest/tar_index.csv` lists all uploaded tar paths and video counts. - `tiktok/preview/stats.json` has global duration / bitrate / resolution histograms. - Preview samples are a small subset of mp4 files for browsing only. - Median duration is 82.2s; median file size is 15.62 MB.