fhswf/dgs-pose

---
license: mit
license_name: fh-internal
license_link: LICENSE
pretty_name: >-
Datenset zur Implementierung von Einzelgebärdenerkennung für Gebärden der
Deutschen Gebärdensprache
task_categories:
- video-classification
language:
- gsg
tags:
- german-sign-language
- sign-language
- pose-estimation
- single-sign
- gesture-recognition
- sign-language-recognition
---

# Datenset zur Implementierung von Einzelgebärdenerkennung für Gebärden der Deutschen Gebärdensprache

## Dataset Summary

This dataset consists of short video clips and pose estimation outputs capturing individual signs from German Sign Language (Deutsche Gebärdensprache, DGS). It is designed for training and evaluating models on isolated sign recognition tasks. The dataset includes both raw video data and `.pev` files, which adhere to the `pose-estimation-recognition-utils` V1.0 format.

## Contents

- Videos of isolated signs in two versions:
- A standard 5-second version
- A trimmed version (≥2 seconds) focused on the signing motion, optionally stretched to meet training length requirements
- Pose estimation files (`.pev`) in standardized format
- from rtmlib by FHSWF (skeleton)
- old data from created with mediapipe (skeleton-mediapipe)

The dataset merges two previously separate sources, which results in some duplicated glosses with different IDs.

Here is the list of all signs with their associated IDs (merged by meaning):

| Sign | IDs |
|------------|----------------------------------|
| a | 39 |
| b | 40 |
| c | 41 |
| d | 42 |
| e | 43 |
| f | 44 |
| g | 45 |
| h | 46 |
| i | 47 |
| j | 48 |
| k | 49 |
| l | 50 |
| m | 51 |
| n | 52 |
| o | 53 |
| p | 54 |
| q | 55 |
| r | 56 |
| s | 57 |
| t | 58 |
| u | 59 |
| v | 60 |
| w | 61 |
| x | 62 |
| y | 63 |
| z | 64 |
| ä | 65 |
| ö | 66 |
| ü | 67 |
| ß | 68 |
| sch | 69 |
| ch | 32 |
| aa | 33 |
| ee | 34 |
| verstanden | 35 |
| nochmal | 36 |
| weiter | 37 |
| Ende | 38 |

## Speakers and Splits

- Speakers 2, 4, 6, and 11 are intended for **validation and testing** only.
- Speaker identity mapping:
- 1 ≈ 2
- 3 ≈ 4
- 5 ≈ 6
- Not all signs are performed by every speaker.

## Annotation Quality

All annotations and video data have been **reviewed and verified by a fluent signer** of German Sign Language.

## Applications

- Video classification (Isolated sign recognition)
- Pose-based recognition using 2D body keypoints
- Data-efficient training and evaluation of models for gesture/action recognition in sign language

## License

This project is published under MIT-license.