metadata
dataset_info:
features:
- name: image
dtype: image
- name: conditioning_image
dtype: image
- name: text
dtype: string
splits:
- name: train
num_bytes: 111989279184.95
num_examples: 507050
download_size: 112032639870
dataset_size: 111989279184.95
Dataset Card for "hagrid-mediapipe-hands"
This dataset is designed to train a ControlNet with human hands. It includes hand landmarks detected by MediaPipe(for more information refer to: https://developers.google.com/mediapipe/solutions/vision/hand_landmarker). The source image data is from HaGRID dataset and we use a modified version from Kaggle(https://www.kaggle.com/datasets/innominate817/hagrid-classification-512p) to build this dataset. There are 507050 data samples in total and the image resolution is 512x512.