Embedding a GIF into your blog post is often a hilarious way to illustrate your reactions, but the GIFs themselves can be low in resolution, like text running across an image that’s too blurry to read. Gfycat is offering an AI-powered solution to improve GIF quality with machine learning, according to a press release.
“Human curation doesn’t take into consideration people’s tastes and mood, so we’ve created our own AI tools,” CEO Richard Rabbat told The Verge in a statement.
Because the company loves cats, it has named each of its three parts of the solution after a feline. The first, called Project Angora, involves searching online for a GIF video source, then swapping out the GIF with a higher-res version. The company says it uploads 15,000 GIFs a day onto its platform and has created 2 million high-quality GIFs so far.
Next, under Project Maru, Gfycat will use facial recognition to identify GIFs even when the user who uploads them hasn’t provided tags. The company says it’s trained the system with a larger dataset to differentiate between individual faces, telling apart Asian actresses like Lucy Liu and Constance Wu — something it was unable to do before.
In the event that a GIF has grainy text, Gfycat has Project Felix to better read and understand illegible captions. Felix uses statistical analysis to guess where text is likely to appear in a GIF and then feeds a binary image into an optical character recognition system to create digital text.
By Shannon Liao at theverge