Ronald Luc
  • Projects
  • Machine learning
  • Tags
  • About

    Pix2pix

    Day ‒> night ; edges ‒> shoes ; landmarks ‒> face

    less than 1 minute read

    Photo credit: Ronald Luc
    Ronald Luc

    Ronald Luc

    Python & Cpp ML developer

    • San Francisco / Czechia
    • Twitter
    • LinkedIn
    • GitHub
    • Stack Overflow
    • YouTube
    Source: own work

    The pix2pix architecture is great for generating images from edges or 2D labels. It can easily change day into night, winter to summer etc. It’s limitations are high RAM demand and hard-to-train GAN model.

    Source: Image-to-Image Translation with Conditional Adversarial Networks

    Full paper: https://arxiv.org/pdf/1611.07004.pdf

    Tags: Architecture, GAN, Medium, Neural networks

    Categories: AI

    Updated: September 10, 2018

    Share on

    Twitter Facebook Google+ LinkedIn
    Previous Next

    You may also enjoy

    50 GPT Image 2.5 prompts, JSON and prose, with notes

    Sep 26, 2026 · 8 minute read

    150 community prompts ‒> 50 picked ‒> notes on what each part does, and 60 test images

    A supply chain conveyor video from one prompt

    Sep 24, 2026 · 8 minute read

    One prompt ‒> 6k lines of Blender Python ‒> 20 s video. Supplier, assembly plant, distributor and a pile of gold coins.

    A 3D periodic table pencil stand from one prompt

    Sep 24, 2026 · 7 minute read

    One prompt ‒> Blender model ‒> printable pencil stand. John Denker’s 3D periodic table with 3 mm walls and a floor.

    Counting green roofs in Brno from aerial photos

    Sep 10, 2026 · 15 minute read

    Aerial photo ‒> roof pixels ‒> plants or not, 2 detectors, 4 years, 1 honest count of Brno’s green roofs.

    • Follow:
    • Twitter
    • GitHub
    • Feed
    © 2026 Ronald Luc. Powered by Jekyll & Minimal Mistakes.