Fonts

Yunhao Ge, Sami Abu-El-Haija, Gan Xin, Laurent Itti

University of Southern California

Overview

Fonts is a computer-generated RGB image datasets. Each image, with 128 * 128 pixels, contains an alphabet letter rendered using 5 independent generating attributes: letter identity, size, font color, background color and font. Below figure shows some samples: in each row, we keep all attributes values the same but vary one attribute value. Attribute details are shown in Table 1. The dataset contains all possible combinations of these attributes, totaling to 1560000 images. Generating attributes for all images are contained within the dataset.

Our primary motive for creating the Fonts dataset, is that it allows fast testing and idea iteration, on disentangled representation learning and zero-shot synthesis.

Fonts_v2 adds simple words in identity attributes, and add three new attributes: position, rotation nd texture. All code are released!

Download and Source code

You can download the Fonts dataset here Download link
(version 1 contains 1.56 million images with nearly 2.5 GB)

The code is released (version1 and version2)! You can freely modify attribute values and create more attributes. [Code]

If you use our dataset or code, please cite the following paper, thanks!

BibTeX


@inproceedings{ge2021zeroshot,
  title={Zero-shot Synthesis with Group-Supervised Learning},
  author={Yunhao Ge and Sami Abu-El-Haija and Gan Xin and Laurent Itti},
  booktitle={International Conference on Learning Representations},
  year={2021},
  url={https://openreview.net/forum?id=8wqCDnBmnrT}
}

Related Work

Zero-shot Synthesis with Group-Supervised Learning
Yunhao Ge, Sami Abu-El-Haija, Gan Xin and Laurent Itti
arXiv:2009.06586, 2020.

[Paper] [Code] [Webpage] [Talk Video] [Fonts Dataset]