It is designed for easy data handling and fast integration with popular deep learning frameworks. We have defined benchmark tasks with baseline scores to asses your model's performance. Have fun!
With ArtSight, the integrated exploration tool you can easily navigate by color (hue, saturation and luminescence levels), creation period, genre, types, objects and so much more. It offers a glipmse into the vast body of data that is OmniArt.
Predefined subsets of data are neatly packed and available for download from this website. Just click on the subset you need and a archive containing the processed images, metadata and data loaders will find their way on your local drive. Just extract and start having fun!
We know how cumbersome the task of writing dataloaders is. You never really get over this part of research. So to help ease the pain we offer special data loaders for popular deep learning frameworks like PyTorch, Keras, Tensorflow. Thanks us later!
Assembling all this data provided the means to do some really interesting experiments for which we report baseline scores. We are curious to see how much we can improve as a comunity. Use our evaluation server and get to the top of the ranked list! Good luck!
Train your own artist attribution expert.
Is it a sketch or a drawing?
Determine the Genre of any painting.
Estimate when an artwork was created.
But where did it come from?
What's my style?
Pinpoint the creation year