image image | label class label |
|---|---|
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 | |
0-K57q5o3dn4#000050#000481 |
This is a ported version of maxin-cn's Taichi-HD dataset for a more convenient use. For each data row, the image is the video frame and the label is the name of this video clip, i.e., frames from the same clip are under the same label.
The dataset is uploaded by following these steps:
- follow the intructions to download the original dataset and unzip;
- create a folder-based Huggingface dataset;
- upload via
push_to_hub.
An investigation on data availability
Note that even it has not been mentioned in maxin-cn's repo, I assume this dataset was originally from this paper: https://github.com/AliaksandrSiarohin/first-order-model.
According to the metadata, the original Taichi-HD dataset has 3049 videos for training and 285 for testing. These 3334 video clips are originally from 252 and 28 YouTube source videos.
However, the maxin-cn's repo has only 2818 training videos, 231 fewer than the full size, while testing videos are complete. As per discussions in Taichi-HD dataset's Github repo, it is very likely due to that many source videos are private, removed, copyright-blocked, or unavailable.
As my investigation, these missing 231 clips attribute to 23 source YouTube videos. On the day I conducted the investigation, the states of these 23 videos are as below.
Note that PKE4yfDESa0 and _L745tFFmCQ seem still publicly available. I assume that on the day maxin-cn built the dataset, they were somehow unavailable or otherwise missed.
- Downloads last month
- 1,250