Synthetic Line Art Dataset
NOTE: This dataset is outdated. An improved version can be found here: SYNLA+
This dataset is designed to simulate complex line art. Useful for training machine learning models which perform any of the following:
- Super-Resolution/Deblurring
- Denoising
- Artifact removal (de-ringing, non-gaussian degradation, etc.)
- Inpainting
- User-Guided Colorization
- Style Transfer
- And more...
Most line art are licensed and have copyright. This dataset offers an open alternative and is released under MIT license.
Two dataset versions are available, one in color, the other in grayscale. Both datasets contain ~2000 images of size 256x256. All images were generated using images in the folder /Generator_Images
, which is also public, allowing custom generation.
The code used to generate the images is not yet public.