“The main challenge is to run the app on the 6GiB RAM iPhone devices.”
iOS kills any app that passes about 2.8GiB on a 6GiB phone. The full-precision denoiser alone is 3.2GiB. Liu Liu got it to run in about 2GiB with FP16, in-place softmax and careful pacing of GPU work. Every AI company says this needs a data center. One guy put it in your pocket at about a minute an image.