Pix2Pix AI wants to transform your doodles and sketches into something remotely resembling a photo of what you were drawing. Sounds confusing but it really isn’t…
Doodling is something most people can do. The Doodler does not have to be a budding Picasso and there are no set requirements for how good a doodle should be. We sent the link out to a few people here at gadgethelpline, the results have been added below for your enjoyment.
This AI will give you instant results and satisfaction. In fact, this generated many hysterical laughs across the office as the results came in.
[ngg_images source=”galleries” container_ids=”4″ display_type=”photocrati-nextgen_basic_slideshow” gallery_width=”600″ gallery_height=”400″ cycle_effect=”fade” cycle_interval=”10″ show_thumbnail_link=”0″ thumbnail_link_text=”[Show thumbnails]” order_by=”sortorder” order_direction=”ASC” returns=”included” maximum_entity_count=”500″]
Pix2Pix uses a generative adversarial network (GAN). This is a method for training AI to work towards a solution to a problem. It does this by judging accuracy of the solution then changing it several times until it reaches a result. You can imagine this like a group of people sketching the same thing then throwing away everything but the closest to the original. This is how Pix2Pix can generate almost realistic images from a simple sketch.
Although Gan’s normally process large data-sets. In this instance the GAN is handling a very small amount of data to work out what you are drawing. It basically copies what you draw then fills in the gaps to make a more complete picture.
The link does not seem to work very well with smartphones unless you can enable desktop view.