PoseText_README.md
January 9, 2024 ยท View on GitHub
Pose Annotation Explainations:
- Please check the table below, please make sure the same frame numbers after your preprocess.
- Each json file contains all frames annotations of one motion and goes by {frame_idx, annotations}.
| Dataset | Clip Number | Frame Number |
|---|---|---|
| aist++ | 1470 | 340928 |
| animation | 329 | 38136 |
| dance | 163 | 36078 |
| egobody | 980 | 438956 |
| fitness | 16730 | 3584563 |
| game_motion | 10217 | 1120002 |
| GRAB | 1335 | 406264 |
| HAA500 | 5231 | 311592 |
| humanml | 26292 | 3579846 |
| humman | 744 | 104981 |
| idea400 | 12513 | 2594858 |
| kungfu | 1040 | 257764 |
| music | 3565 | 876748 |
| perform | 475 | 102522 |
Our basic usage (only random sample one frame pose description):
with open(body_text_name + '.json', 'r') as body_f:
body_dict = json.load(body_f)
with open(body_text_name + '.json', 'r') as hand_f:
hand_dict = json.load(hand_f)
select_frame = random.randint(0, len(hand_dict)-1)
hand_frame_text, body_frame_text = hand_dict[str(select_frame)], body_dict[str(select_frame)]