Load Flickr Faces HQ (FFHQ) dataset in Python fast with one line of code. Dataset of 70,000+ faces in HQ. Stream FFHQ while training models in PyTorch & TensorFlow.
Visualization of the FFHQ Dataset on the Activeloop Platform
FFHQ (Flickr-Faces-HQ) Dataset
What is FFHQ Dataset?
The FFHQ Dataset (Flickr-Faces-HQ) dataset is a high quality image set containing 70,000 PNG images at 1024x1024 resolution, its been put a considerable effort to include as many attributes as possible and variations on these, so expect a wide range of different age, and ethinicty.
Importantly, this dataset is not intended for, and should not be used for, development or improvement of facial recognition technologies.
Download FFHQ Dataset in Python
Instead of downloading the FFHQ dataset in Python, you can effortlessly load it in Python via our open-source package Hub with just one line of code
Load FFHQ Dataset in Python
import hub
ds = hub.load("hub://activeloop/ffhq")
FFHQ Dataset Structure
FFHQ Data Fields
face_landmarks: a int32 tensor containing a 204 coco points
image_102: a uint8 tensor containing a 1024 x 1024 png image
image_128: a uint8 tensor containing a 128 x 128 png image
images_metadata: a str tensor containing the 70,000 image metadata
face_landmarks: a int32 tensor containing 204 coco points
face_quad: a float32 tensor containing 4 generic objects
face_rect: a int32 tensor containing 1 bbox object
image: a uint8 tensor containing a 1062 x 1072 compressed png image
FFHQ Data Splits
While FFHQ dataset is not explicitly split between a training and validation set, it is customary to use the first 60,000 images as a training set and the remaining 10,000 for validation. You can achieve that in the following way:
Train a model on FFHQ dataset with TensorFlow in Python
dataloader = ds.tensorflow()
FFHQ Dataset Creation
Data collection and Normalization of images
These images were collected from Flickr under a permissive license, they were cropped and aligned using dlib and various filters were used to prune the set. Finally, Amazon Mechanical Turk was used to remove the occasional statues, paintings, or photos of photos.
Paper: T. Karras, S. Laine and T. Aila, "A Style-Based Generator Architecture for Generative Adversarial Networks," 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 4396-4405, doi: 10.1109/CVPR.2019.00453.
The FFHQ dataset(Flickr-Faces-HQ) is a popular high quality image dataset scraped from twitter. FFHQ dataset provides 70,000 PNG images of a wide range of people of different features, these images are provided with a resolution of 1024 x 1024.
What is the FFHQ dataset used for?
FFHQ is used as a means for the benchmarking of generative adversarial networks models, but the wide range of age and other features allows for multiple uses on this dataset such as face aging, face generation, and metric learning.
How to download the FFHQ dataset in Python?
You can load the FFHQ dataset fast with one line of code using the open-source package Activeloop Hub in Python. See detailed instructions on how to load the FFHQ dataset.
How can i use FFHQ dataset in Pytorch or TensorFlow?