Awesome

FIGR-8

The FIGR-8 database is a dataset containing 17,375 classes of 1,548,256 images representing pictograms, ideograms, icons, emoticons or object or conception depictions. Its aim is to set a benchmark for Few-shot Image Generation tasks, albeit not being limited to it. Each image is represented by 192x192 pixels with grayscale value of 0-255. Classes are not balanced (they do not all contain the same number of elements), but they all do contain at the very least 8 images.

The main contribution of this dataset to the community is the abundance of image classes, which is valuable for tasks in which encompassing an abundance of concepts can be useful or even necessary. This was the case in our paper, FIGR: Few-shot Image Generation with Reptile.

The dataset is readily available for people who want to give a shot at few-shot image generation techniques, while spending minimal effort on gathering a full-size database. We also encourage this dataset to enable reproduction efforts from the community.

Contents

• Data/
  • All images are contained in subfolders inside Data folder.
• data.csv with the following information in order (see below for example):
  • Category number. All items from the same category name have the same category number. One category number per category name.
  • Category name. Also called "class". Represents what is depicted in the image.
  • Path. Represented as Category name/image_id. All images are in 192x192 with .png extension.
  • Artist. The artist who designed the logo, the icon or the pictogram.
  • License. The license type which the image is subject to. Refer to section License for more information.

Sample from data.csv:

Alternate Download

Alternatively, you can download the dataset from the FIGR-8 Google Drive, or get the .torrent link from Academic Torrents.

SVG images

If you prefer to work with .SVG (vector graphics) images, you can download an alternate version of the images from the FIGR-8-SVG Google Drive, from the FIGR-8-SVG github repository, or from Academic Torrents.

License

Most images in the dataset have been licensed under a Creative Commons License by their author, indicating that their reproduction on any material intended to be sold or to be made profit from is strictly prohibited. However, use in which the author's name is indicated is permitted. More details can be found here.

The dataset itself (FIGR-8) is protected under the MIT License.

Acknowledgement

Images were gathered from The Nounji App, available on the App Store.

If you use this database for your own projects, please consider citing the following paper:

@article{DBLP:journals/corr/abs-1901-02199,
  author    = {Louis Clou{\^{a}}tre and
	       Marc Demers},
  title     = {{FIGR:} Few-shot Image Generation with Reptile},
  journal   = {CoRR},
  volume    = {abs/1901.02199},
  year      = {2019},
  url       = {http://arxiv.org/abs/1901.02199},
  archivePrefix = {arXiv},
  eprint    = {1901.02199},
  timestamp = {Thu, 31 Jan 2019 13:52:49 +0100},
  biburl    = {https://dblp.org/rec/bib/journals/corr/abs-1901-02199},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}