datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
google-landmarkshow2sign-asl-landmarks
How2Sign ASL Landmarks
Sentence-level MediaPipe landmark cache for the How2Sign ASL translation dataset. This dataset stores preprocessed landmark arrays only, not MP4 clips.
The rows come from the local How2Sign realigned CSV files and front-view raw videos. Each successful row is one sentence segment sampled at a 25 FPS cap and stored in compressed NPZ shards.
Why This Dataset Exists
The public How2Sign data has known alignment/completeness issues: some text rows do not… See the full description on the dataset page: https://huggingface.co/datasets/martinctl/how2sign-asl-landmarks.how2sign-landmarks-front-raw-parquet
Split
Count
train
31047
validation
1739
test
2343
Parquet-shared front mediapipe data from PSewmuthu/How2Sign_Holistic
import cv2
import numpy as np
import tempfile
import os
from datasets import load_dataset
from IPython.display import Video, display
# 1. Initialize Streams
print("Initializing streams...")
landmark_stream = load_dataset("bdanko/how2sign-landmarks-front-raw-parquet", split="train", streaming=True)
rgb_stream =… See the full description on the dataset page: https://huggingface.co/datasets/bdanko/how2sign-landmarks-front-raw-parquet.Level-5-Cultures-Landmarks-1vtl-speech-landmarks
VTL Speech Landmarks Dataset
Articulatory speech synthesis dataset with acoustic landmarks, generated using VocalTractLab (VTL).
Dataset Description
This dataset contains synthesized speech for 117,497 English words from the CMU Pronouncing Dictionary, generated with two speakers (male and female). Each word includes:
Audio: 48kHz WAV files
Landmarks: Acoustic-phonetic event markers (JSON)
Articulatory data: Full vocal tract trajectories from VTL (JSON)
Speakers… See the full description on the dataset page: https://huggingface.co/datasets/mcamara/vtl-speech-landmarks.google_landmarks_places
Google Landmarks places
Google Landmarks is a great dataset, but it lacks geospatial information about the places. This dataset fills
this gap by providing latitude and longitude for each landmark. The dataset also contains the name of the landmark from OpenStreetMap
and information about the country, the province/state, and the city/village where the landmark is located. This information was collected from OSM
via Nominatim.
nsaka-complete-landmarkslandmarks
Dataset Card for "landmarks"
More Information needed
google-landmarks-v2-miniOptimized_Video_Facial_Landmarks
Dataset Card for 478-Point Normalized 3D Facial Landmark Dataset
Dataset Description
This dataset provides pre-extracted, normalized 3D facial landmark features derived from the Video Emotion dataset. It is optimized for efficient training of emotion recognition and facial analysis models, bypassing the need to process large raw video files.
License: The extracted feature data in this Parquet file is licensed under Apache 2.0. Note that the original source video files may… See the full description on the dataset page: https://huggingface.co/datasets/PSewmuthu/Optimized_Video_Facial_Landmarks.gym-exercise-landmarksEmotion_Video_Facial_Landmarks
Dataset Card for 478-Point Normalized 3D Facial Landmark Dataset
Dataset Description
This dataset provides pre-extracted, normalized 3D facial landmark features derived from the Video Emotion dataset. It is optimized for efficient training of emotion recognition and facial analysis models, bypassing the need to process large raw video files.
License: The extracted feature data in this CSV file is licensed under Apache 2.0. Note that the original source video files may… See the full description on the dataset page: https://huggingface.co/datasets/PSewmuthu/Emotion_Video_Facial_Landmarks.how2sign-landmarks-curated-normalizedA specially selected set of 96 mediapipe coordinates from Mediapipe Holistic.
import numpy as np
from datasets import load_dataset
dataset_name = "bdanko/how2sign-landmarks-curated-normalized"
print(f"Loading dataset {dataset_name}...")
dataset = load_dataset(dataset_name, split="test", streaming=True)
# Get a sample
sample = next(iter(dataset))
video_id = sample['video_id']
sentence = sample['sentence']
features_raw = sample['features']
shape = sample['shape']
# Decode features
landmarks… See the full description on the dataset page: https://huggingface.co/datasets/bdanko/how2sign-landmarks-curated-normalized.google_landmarks_photos
Dataset Card for "google_landmarks_photos"
More Information needed
tunisian-landmarks-captions
🇹🇳 Tunisian Landmarks Image-Caption Dataset
An image-caption dataset documenting Tunisia's rich geographical, historical, and architectural heritage. The dataset pair high-resolution photographs of iconic Tunisian sites with detailed, human-curated descriptions highlighting their cultural significance, geographical features, and architectural styles.
Dataset Details
Dataset Description
This dataset is built to support vision-language tasks such as… See the full description on the dataset page: https://huggingface.co/datasets/firastlili/tunisian-landmarks-captions.Level-4-Cultures-Landmarks-1CASL-W60-Landmarks
CASL-W60-64-frames Landmarks
license: mit
task_categories:
- video-classification
language:
- en
tags:
- sign-language
- mediapipe
- skeleton
- Central-African-Sign-Language
pretty_name: CASL-W60 Landmarks
size_categories:
- 1K<n<10K
Dataset Card for CASL-W60 Landmarks
This dataset contains preprocessed holistic landmarks for Central African Sign Language (CASL). It is specifically designed for training sequence-based models like Transformers or LSTMs for Sign… See the full description on the dataset page: https://huggingface.co/datasets/luciayen/CASL-W60-Landmarks.slovo-full-landmarks
Slovo landmarks base
Base landmarks dataset built from full Slovo videos with MediaPipe Tasks HolisticLandmarker.
Emotion_Video_Facial_Landmarks
Dataset Card for 478-Point Normalized 3D Facial Landmark Dataset
Dataset Description
This dataset provides pre-extracted, normalized 3D facial landmark features derived from the Video Emotion dataset. It is optimized for efficient training of emotion recognition and facial analysis models, bypassing the need to process large raw video files.
License: The extracted feature data in this CSV file is licensed under Apache 2.0. Note that the original source video files may… See the full description on the dataset page: https://huggingface.co/datasets/mac26/Emotion_Video_Facial_Landmarks.how2sign-clips-landmarksThis dataset is from:
@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}
}
Also see the offical how2sign dataset website:… See the full description on the dataset page: https://huggingface.co/datasets/tim1047/how2sign-clips-landmarks.breast-mri-sagitall-landmarks
Breast MRI sagittal landmarks
Breast-region masks and nipple landmarks for 35 sagittal, unilateral breast DCE-MRI
exams.
This is a subset of the sagittal exams from the TCIA QIN-BREAST collection:
https://www.cancerimagingarchive.net/collection/qin-breast/. The images are redistributed
unchanged from that collection under its CC BY 3.0 licence; the masks are new.
Sagittal breast MRI is under-represented in public landmark data. Models trained on axial
bilateral exams are known… See the full description on the dataset page: https://huggingface.co/datasets/h2thez3/breast-mri-sagitall-landmarks.Taichi-HD-Landmarks
A subset of Taichi-HD dataset with estimated body landmarks, ported from https://github.com/AliaksandrSiarohin/motion-cosegmentation.
The protocol of dataset creation can be found at the original paper: https://arxiv.org/abs/2004.03234
We backup the raw images (./raw_assets) and landmarks (./landmark) in this repo.
isl_landmarks_featureswelsh-speech-landmarks
Welsh Speech Dataset - Facial Landmarks
68-point facial landmarks (ibug68 template) from the Welsh Speech Dataset.
Contents
Facial landmarks for every frame
68 3D points per frame (x, y, z coordinates)
Format: Parquet
Manual annotation using ibug68 template
Format
The landmarks.parquet file contains:
Column
Description
speaker_id
Speaker identifier (1-33)
phrase_id
Phrase identifier (1-10)
frame_id
Frame identifier (e.g., "001", "002")… See the full description on the dataset page: https://huggingface.co/datasets/arvinsingh/welsh-speech-landmarks.bio_landmarks_zh_gemma-2-9b-itwlasl-upper-arm-pose-landmarks
WLASL Pose Landmarks (MediaPipe Lite)
This dataset contains 3D pose landmarks extracted from the WLASL (World Level American Sign Language) video dataset using Google MediaPipe. It is designed to facilitate lightweight sign language recognition models by providing pre-computed skeletal data instead of raw video pixels.
Supported Tasks
Sign Language Recognition (SLR)
Skeleton-based Action Recognition
Keypoint Analysis
Dataset Structure
The dataset contains… See the full description on the dataset page: https://huggingface.co/datasets/Kibalama/wlasl-upper-arm-pose-landmarks.europe-landmarks-classification
Europe Landmarks Classification
This dataset was created as part of my master's thesis. It's comprised of several thousand images of landmarks in European cities. To comply with image licenses, this dataset also includes. csv attribution file with information about the author of the image, the license under which it was released, and a link to the original image on Wikipedia Commons.
Dataset Structure
The dataset is comprised of 50 classes of selected landmarks in… See the full description on the dataset page: https://huggingface.co/datasets/SablikJan/europe-landmarks-classification.asl-static-landmarks-v1
ASL Static Letter Recognizer (Real-Time)
Real-time ASL fingerspelling recognizer built with MediaPipe + TensorFlow.
This project recognizes 24 static ASL letters from webcam input.
✅ Included: static letters (e.g., A, B, C, ...)
❌ Excluded: J, Z (motion-based letters)
Current model status
Final static model: models/asl_static_model.h5
Label encoder: models/static_label_encoder.pkl
Input features: 86 engineered features from 21 hand landmarks
Reported accuracy: 92.74%… See the full description on the dataset page: https://huggingface.co/datasets/Siruyy/asl-static-landmarks-v1.ChicagoFSWild-Landmarksaahq-face-landmarks-30k