SNGFaces Dataset

SNGFaces is an image dataset of human faces extracted from incredibly high resolution scans (often over 0.5GB per scan) of oil paintings, heavily inspired by NVIDIA's MetFaces dataset.

The dataset consists of 621 high-quality PNG images at 2048×2048 resolution and a superset of 644 high-quality PNG images at 1024×1024.

The images were downloaded via the on-line catalog of artworks developed by lab.SNG: the digital R&D lab of the Slovak National Gallery (SNG), and automatically aligned and cropped using original MetFaces preprocessing script. Only images under public domain license were collected.

Licenses

The source images are made available under the Public Domain Mark 1.0. Please read here for more information about source's license policies.

The dataset itself (including JSON metadata, processed images, and documentation) is made available under Creative Commons BY-NC 2.0. You can use, redistribute, and adapt it for non-commercial purposes, as long as you (a) give appropriate credit by referencing this repository, and (b) indicate any changes that you've made.

Overview

All data is hosted on Google Drive:

Path	Size	Files	Format	Description
sngfaces-dataset	6.0 GB	2023		Main folder
├ sngfaces.json	0.3 MB	1	JSON	Image metadata
├ unprocessed_previews	0.6 GB	757	PNG	Original unprocessed preview images at 800×(shorter size) resolution
├ processed/2048	4.3 GB	621	PNG	Aligned and cropped images at 2048×2048 resolution
└ processed/1024	1.1 GB	644	PNG	Aligned and cropped images at 1024×1024 resolution

Reproducing the dataset

Contrary to MetFaces, SNGFaces images can not be reproduced since the original full resolution images are not provided as their total size is well over 100GB.

Metadata

The sngfaces.json file contains the following information for each image:

{
   "SVK:SNG.O_4903":{
      "properties":{
         "id":"SVK:SNG.O_4903",                            # Inventory number of the paitning
         "author":"Kálmán Rudnyánszky",
         "title":"Portrait of Baron Gejza Mednyánszky",
         "date":"1855",
         "institution":"Slovenská národná galéria, SNG",
         "work type":"painting",
         "genre":"portrait",
         "material":"canvas",
         "technique":"oil"
      },
      "image_2048_md5":"9ecdeb74d5ab4d05ec3ab15ee49526c6", # MD5 checksum of the 2048x2048 image
      "image_1024_md5":"4bd63fc9a473c4d72a4c0778bf36cb74"  # MD5 checksum of the 1048x1048 image
   },
   ...
}

Differences between SNGFaces and MetFaces

• SNGFaces source images are incredibly high quality color scans (majority has well over 10,000 pixels per image side and 100MBs of file size) coordinated by professional from SNG's Department of Digital Collections and Services. As such SNGFaces contain unscaled images at pure 2048×2048 resolution. On the contrary, some amount of MetFaces images are upscaled greyscale photographs at 1024×1024 resolution.
• SNGFaces is exclusively made of portraits of oil paintings produced on canvas while MetFaces also includes other forms of art, such as sculptures.
• As noted in Reproducing the dataset section, SNGFaces can not be reproduced contrary to MetFaces which include all source images and the processing script.