Digital face aging with neural network

Disney Research demonstrates their use of neural networks to seamlessly age and de-age someone’s face across a continuous range. Sure what is real anymore anyway.

Tags: , , ,

AI-generated movie posters

Noah Veltman fed an AI movie descriptions and made it generate images. The results are in quiz form so that you can guess the movies. I would give myself a poor rating for guessing the movies, but once you see the answer, you’re like oh yeah of course.

Veltman used VQGAN+CLIP, which you can find out more about here.

Tags: , , ,

Converting Minecraft worlds to photorealistic ones using neural networks

Researchers from NVIDIA and Cornell University made GANcraft:

GANcraft aims at solving the world-to-world translation problem. Given a semantically labeled block world such as those from the popular game Minecraft, GANcraft is able to convert it to a new world which shares the same layout but with added photorealism. The new world can then be rendered from arbitrary viewpoints to produce images and videos that are both view-consistent and photorealistic.

This is impressive, but what amazes me more is that Minecraft is still very much a thing after all these years.

Tags: , ,

AI-generated candy hearts

Continuing her annual tradition, Janelle Shane trained various AI models to generate two-word-all-caps love messages for those chalky Valentine’s Day candy hearts. So deep. So profound.

See also Shane’s experiment with generating hearts for somewhat creepy results, as AI often likes to do.

Tags: , , ,

Neural network creates images from text

OpenAI trained a neural network that they call DALL·E with a dataset of text and image pairs. So now the neural network can take text input and output random combinations of descriptors and objects, like a purse in the style of Rubik’s cube or a teapot imitating Pikachu.

Tags: , , ,

Neural network generates convincing songs by famous singers

Jukebox from OpenAI is a generative model that makes music in the same styles as many artists you’ll probably recognize:

To train this model, we crawled the web to curate a new dataset of 1.2 million songs (600,000 of which are in English), paired with the corresponding lyrics and metadata from LyricWiki. The metadata includes artist, album genre, and year of the songs, along with common moods or playlist keywords associated with each song. We train on 32-bit, 44.1 kHz raw audio, and perform data augmentation by randomly downmixing the right and left channels to produce mono audio.

A lot of the time, generative music sounds artificial and mechanical, but these results are pretty convincing. I mean you can still tell it’s not from the artist, but many of the examples are listenable.

OpenAI also published the code.

Tags: , , ,

Text-to-speech models trained on celebrity voices

The Vocal Synthesis channel on YouTube trains text-to-speech models using publicly available celebrity voices. Then using this new computer-generated voice, the celebrities “recite” various scripts. For example, the above is Jay-Z rapping the “To be, or not to be” soliloquy from Hamlet, but it’s not him.

Find out more about the voice generation here, which was developed in 2017. Maybe more interesting, Jay-Z recently filed a copyright claim against the videos.

Tags: , , ,

Neural networks to generate music

Kyle McDonald describes some of the history and current research on using algorithms to generate music. On how David Cope incorporated Markov chains to aid in his work:

In 1981 David Cope began working with algorithmic composition to solve his writers block. He combined Markov chains and other techniques (musical grammars and combinatorics) into a semi-automatic system he calls Experiments in Musical Intelligence, or Emmy. David cites Iannis Xenakis and Lejaren Hiller (Illiac Suite 1955, Experimental Music 1959) as early inspirations, and he describes Emmy in papers, patents, and even source code on GitHub. Emmy is most famous for learning from and imitating other composers.

I expected samples to sound robotic and unnatural, and some are, but some are quite pleasant to listen to.

Tags: ,

Algorithmic art shows what the machine sees

Tom White is an artist who uses neural networks to draw abstract pictures of objects. What looks blobby and fuzzy to us looks more concrete to the machine.

James Vincent for The Verge:

That “voice” is actually a series of algorithms that White has dubbed his “Perception Engines.” They take the data that machine vision algorithms are trained on — databases of thousands of pictures of objects — and distill it into abstract shapes. These shapes are then fed back into the same algorithms to see if they’re recognized. If not, the image is tweaked and sent back, again and again, until it is. It’s a trial and error process that essentially ends up reverse-engineering the algorithm’s understanding of the world.

Tags: ,

Neural networks to communicate with Alexa devices using sign language

Many have found Amazon’s Alexa devices to be helpful in their homes, but if you can’t physically speak, it’s a challenge to communicate with these things. So, Abhishek Singh used TensorFlow to train a program to recognize sign language and communicate with Alexa without voice.

Nice.

Tags: , , ,