---
pretty_name: How2Sign Holistic
language: en
license:
  - mit
tags:
  - sign-language
  - asl
  - mediapipe
  - holistic
  - pose-landmarks
  - hand-landmarks
  - face-landmarks
  - gesture-recognition
  - sequence-modeling
  - time-series
  - computer-vision
  - deep-learning
source_datasets:
  - Duarte_CVPR2021/How2Sign
task_categories:
  - feature-extraction
  - translation
task_ids:
  - pose-estimation
  - conversational
citation:
  - "@inproceedings{Duarte_CVPR2021,
    title={{How2Sign: A Large-scale Multimodal Dataset for Continuous American Sign Language}},
    author={Duarte, Amanda and Palaskar, Shruti and Ventura, Lucas and Ghadiyaram, Deepti and DeHaan, Kenneth and
    Metze, Florian and Torres, Jordi and Giro-i-Nieto, Xavier},
    booktitle={Conference on Computer Vision and Pattern Recognition (CVPR)},
    year={2021}
    }"
  - "@misc{MediaPipe,
    title={MediaPipe},
    author={Google Inc.},
    year={2020},
    url={https://mediapipe.dev/}
    }"
---

# How2Sign Holistic

### Mediapipe Holistic Landmark Features Extracted from the How2Sign ASL Dataset

## Overview

**How2Sign Holistic** is a curated dataset providing frame-level Mediapipe Holistic landmarks extracted from the full How2Sign American Sign Language corpus. Each sentence-level video clip has pose, face, and hand landmark sequences stored as `.npy` files.

This dataset is designed to support research in:

- ASL recognition and translation
- Pose-based sign generation
- Sequence and time-series modeling
- Gesture understanding
- Multiview motion analysis

## Base Directory

**`how2sign_holistic_features/`** is the root folder containing all splits and metadata.

## Sources

The original data comes from the **How2Sign dataset** (Duarte et al., CVPR 2021), a large-scale multimodal American Sign Language dataset sourced from YouTube videos.

## Collection Methodology

- Sentence-level clips were extracted from the original videos according to How2Sign protocol.
- Frame-level landmarks were extracted using **Google Mediapipe Holistic** (pose, face, hands).
- Each clip saved as `.npy` with frontal and side views.
- Metadata CSVs map clips to sentences, start/end timestamps, and video identifiers.
- CSVs can be opened in pandas: `pd.read_csv('filename.csv', sep='\t')`

## Dataset Structure

```
how2sign_holistic_features/
│
├── metadata/                     # Original How2Sign metadata (CSV files)
│   ├── how2sign_realigned_train.csv
│   ├── how2sign_realigned_val.csv
│   ├── how2sign_realigned_test.csv
│   ├── how2sign_train.csv
│   ├── how2sign_val.csv
│   └── how2sign_test.csv
│
├── train/                        # Training split .npy files
│   ├── frontal/
│   │   ├── <VIDEO_ID>_front_holistic.npy
│   │   └── ...
│   └── side/
│       ├── <VIDEO_ID>_side_holistic.npy
│       └── ...
│
├── val/                          # Validation split
│   ├── frontal/
│   └── side/
│
└── test/                         # Test split
    ├── frontal/
    └── side/
```

### Notes

- `.npy` files contain **frame-level Mediapipe Holistic landmarks**.
- Frontal and side views are synchronized.
- Filenames follow: `VIDEO_NAME_START-END-rgb_front/side_holistic.npy`
- Metadata CSVs map clips to video ID, sentence, start/end timestamps, and How2Sign identifiers.

## Citation

If you use this dataset, please cite:

Duarte, A., Palaskar, S., Ventura, L., Ghadiyaram, D., DeHaan, K., Metze, F., Torres, J., & Giro-i-Nieto, X.
**“How2Sign: A Large-scale Multimodal Dataset for Continuous American Sign Language.”**
_Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021._

## Recommended Tags

`ASL`, `Sign Language`, `Mediapipe`, `Holistic`, `Pose Landmarks`, `Hand Landmarks`, `Face Landmarks`, `Keypoints`, `Motion Capture`, `Time Series`, `Gesture Recognition`, `Computer Vision`, `Deep Learning`, `Sequence Modeling`