File size: 2,144 Bytes
ffd25bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9fd219d
ffd25bb
9fd219d
 
 
ffd25bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9fd219d
ffd25bb
 
 
 
 
 
9fd219d
ffd25bb
 
 
9fd219d
ffd25bb
 
9fd219d
 
 
ffd25bb
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
---
license: unknown
task_categories:
- other
language:
- vi
tags:
- sign-language
- vietnamese-sign-language
- pose-estimation
- dwpose
- skeleton
size_categories:
- 10K<n<100K
---

# Multi-VSL (front view) — DWPose skeletons

Whole-body 2D pose keypoints extracted with [DWPose](https://github.com/IDEA-Research/DWPose)
from the front-camera clips of the Multi-VSL Vietnamese Sign Language corpus.

- **28,406** clips, one `.npz` per clip
- Total size: ~4.6 GB
- Laid out as `data/<signer>/<clip>.npz` — 30 signer directories, 628–1,167 clips
  each (HF rejects directories holding more than 10,000 files, so a flat tree is
  not possible here)

## Contents of each `.npz`

| key | shape | dtype | description |
|---|---|---|---|
| `all_xy` | `(T, 128, 2)` | `float16` | keypoint pixel coordinates per frame |
| `all_score` | `(T, 128)` | `float16` | per-keypoint confidence |
| `detected` | `(T,)` | `int8` | 1 if a person was detected in the frame, else 0 |
| `frame_size` | `(2,)` | `int32` | source video frame size |
| `fps` | scalar | `float32` | source video frame rate |

`T` is the number of frames in the clip. The 128 keypoints follow the DWPose /
COCO-WholeBody layout: 17 body + 6 foot + 68 face + 42 hands (21 per hand).

## File naming

```
<session>___<view>_<device>_<signer>_<view>_<order>_<clip_index>.npz
```

e.g. `01_Co-Hien_100-200_1-2-3_0118___center_device10_signer01_center_ord1_100.npz`

## Usage

```python
import numpy as np
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    "Tri1/Multi-VSL-front-skeleton",
    "data/signer01/01_Co-Hien_100-200_1-2-3_0118___center_device10_signer01_center_ord1_100.npz",
    repo_type="dataset",
)
d = np.load(path)
xy, score = d["all_xy"], d["all_score"]   # (T, 128, 2), (T, 128)
```

Download everything, or just one signer:

```python
from huggingface_hub import snapshot_download

snapshot_download("Tri1/Multi-VSL-front-skeleton", repo_type="dataset",
                  local_dir="skeleton")

snapshot_download("Tri1/Multi-VSL-front-skeleton", repo_type="dataset",
                  allow_patterns="data/signer01/*", local_dir="skeleton")
```