Segmented Bob Ross Images

This dataset contains ~250 segmented Bob Ross paintings. Bob Ross was a painter and painting instructor who was on PBS public television for over a decade with his show "The Joy of Painting". Bob Ross is known for his easy-to-learn "wet-on-wet" painting style, the use of vibrant color in his landscape paintings, and for his generally calm, joyous personality.

Despite Bob Ross having passed away in 1995, "The Joy of Painting" continues to run in syndication, and he remains well-known in modern popular culture.

This dataset can be used to build a generative art GAN. For example, I used this dataset to fine-tune a GauGAN model that learns to output "Bob Ross like" images like these:

https://i.imgur.com/A6T6y6o.png" alt="">

It is a suitable starting point for this and other interesting generative art tasks.

Acknowledgements

The Bob Ross image corpus was collected from an unknown source by GitHub user Jared Wilbur. The original image corpus consists of ~400 images. I hand-labelled ~250 of these into nine different classes (see the label key in labels.csv) ranging from "sky" to "mountain", following the label number ontology used by the ADE20K dataset.