All work3rd of 4,000+ teams

AudioVision

Old footage in; color, four times the resolution, and generated sound out. Third of 4,000+ teams worldwide.

AMD Pervasive AI Contest3rd of 4,000+ teamsPyTorchROCm2024Demo filmWriteupGitHub
AudioVision demo film cover
Demo film

What it is

A generative restoration pipeline for historical footage. GAN colorization feeds 4x super-resolution, and a generative audio model scores the result, so a silent black-and-white clip comes back in color with sound.

The engineering

  • Three heavy generative models chained end to end as GPU inference on a single AMD workstation card, on ROCm rather than CUDA.
  • A usable front end over the pipeline, so the demo ran as a product rather than a notebook.
  • I led a team of five, and we demoed it live on AMD's keynote stage.

Next