CelebA with attrs at 128x128 resolution.
The attributes are binary attributes. The dataset is already split into train/test/validation sets.
This dataset has been reduced so there's 160k train samples.
@inproceedings{liu2015faceattributes,
title = {Deep Learning Face Attributes in the Wild},
author = {Liu, Ziwei and Luo, Ping and Wang, Xiaogang and Tang, Xiaoou},
booktitle = {Proceedings of International… See the full description on the dataset page:
https://huggingface.co/datasets/tpremoli/CelebA-attrs-160k.
