Large-scale Celebrity Faces (CelebA)

This dataset contains over 200,000 images of celebrity faces.

The default image size is 128x128, but you can also specify different size during training, e.g.

mantra train sagan --dataset celeba --image-dim 256 256


To import this dataset to your project, run:

mantra import RJT1990/data/celeba

This is a large dataset, so the first time you train with it, it may take a while to version and upload to the cloud. But after you use it once, it should be quick to use.


Example usage in code:

from data.celeba.data import CelebA d = CelebA() print(d.X.__class__) ## <class 'numpy.ndarray'> print(d.X.shape) ## (5680, 128, 128, 3)

Data processing
from mantraml.data.ImageDataset import ImageDataset class CelebA(ImageDataset): # These class variables contain metadata on the Dataset data_name = 'CelebA' data_tags = ['celebrities'] data_image = 'celeba.jpeg' files = ['celeba.tar.gz'] has_labels = False image_dim = (128, 128) # default - can override with command line
All files
celeba / raw /
Last Modified
17 September 2018 2:33 PM
17 September 2018 2:33 PM