— Memo Akten (@memotv) March 7, 2016
Ten hour version with audio on gifsound.
Eurovision’s Epic Sax Guy in the style of Monet’s Poppyfield in Giverny, using state-of-the-art Deep Learning algorithms. #EpicMonetGuy is a reaction to the disposable, sensationalist, instant-gratification-seeking, click-bait ‘art & entertainment’ extravaganza and related debates in the internet age.
There has been an explosion in Deep Learning research recently, allowing images to be synthesised in various novel ways. This includes algorithms which can ‘learn’ the ‘style’ of a particular image (e.g. Monet, Van Gogh etc) and ‘apply’ it to another image, such as a photo. Even though this research is very much in its infancy, it is already technically quite ground-breaking, and many applications of these algorithms have gone viral on social media (and subsequently on general media) with various hashtags such as #Deepdream, #NeuralDoodle, #StyleTransfer etc, (also recently resulting in a Art of Neural Networks exhibition at the Gray Area Foundation in San Francisco). However, the general discussions on media and social media have mostly chosen to focus on ludicrous angles such as ‘Skynet is coming’, ‘computers can be creative for you so now you don’t have to be’, ‘filters to create instant art’ etc. – a very popular tagline being ‘create art by painting like Monet or Van Gogh’. Trying to reduce ‘the creation of art’ to a one-button-click process; off-loading or negating all sense of vision, creativity, intentionality, responsibility or skill to a piece of software; shared via click-bait headlines or throwaway comments on social media or news sites; these discussions simultaneously expose our obsessions for sensationalism, instant gratification and disposable debate. #EpicMonetGuy is a reaction to all of this.
(In case it isn’t obvious, I’m not working on a VR version. That was a nod to the current ‘VR is going to change & save the world’ hype.)
The mentioned research includes:
- A Neural Algorithm of Artistic Style
- Combining Markov Random Fields and Convolutional Neural Networks for Image Synthesis
- Semantic Style Transfer (code)
- Neural Image Analogies
Made with Adam Wentz’s Neural Image Analogies research.