Lj Miranda

ljvmiranda921.github.io · Personal blog · on HN since 2019

Lj Miranda writes about machine learning, Python tooling, and Kubernetes, with explainers on VQGAN and generative art projects.

What HN saysHN commenters link his VQGAN explainer for a technical account of how VQGAN and CLIP generate images, and as a companion to related discussions.
19HN people shared it
3linked it in comments
27stories submitted
617story points
3 months agolast shared
2019people per year2026

Most-shared posts

  1. 1
    The Illustrated Vqgan (2021)
    5 people · 6 points · first shared 2021 · discussion on HN
  2. 2
    Show HN: Generate 8-bit avatars using Cellular Automata
    3 people · 16 points · first shared 2020 · discussion on HN
  3. 3
  4. 4
    How to Use Jupyter Notebooks in 2020
    2 people · 6 points · first shared 2020 · discussion on HN
  5. 5
  6. 6
    What can Avengers Endgame teach us about Git?
    2 people · 4 points · first shared 2021 · 2 comments on HN
  7. 7
    How to Use Notebooks in 2020
    2 people · 3 points · first shared 2020 · discussion on HN
  8. 8
    Why do we need Flask, Celery, and Redis? (2019)
    1 people · 347 points · first shared 2020 · 181 comments on HN
  9. 9
    Thinking with pen and paper
    1 people · 187 points · first shared 2022 · 165 comments on HN
  10. 10
  11. 11
  12. 12
    Train-test split may be doing you a disservice (2022)
    1 people · 2 points · first shared 2024 · discussion on HN
  13. 13
    Burnout Barometer - Mindfulness throughout the day
    1 people · 2 points · first shared 2020 · discussion on HN
  14. 14
    How to Debug
    1 people · 1 points · first shared 2020 · discussion on HN
  15. 15

What HN people wrote when they linked it

lancesells · 2022-08-05
If you have a notebook / sketchbook / journal and it needs to be held open with an object look for another notebook [1]. It seems trivial but that is a poorly designed notebook. 1. https://ljvmiranda921.github.io/assets/png/pen-and-paper/blank.jpg
minimaxir · 2021-09-02
Due to how VQGAN + CLIP works, it can't memorize its inputs in the way language models like GPT-3 do. VQGAN does the generation work, CLIP just says if it's good not, improve the latents, repeat. Here's a good technical writeup: https://ljvmiranda921.github.io/notebook/2021/08/08/clip-vqgan/ And of course, the most-common VQGAN was trained on ImageNet, which likely doesn't have every movie poster as training data. (it could be in CLIP though)
ljvmiranda · 2021-08-10
I also wrote an explainer for VQGAN here: https://ljvmiranda921.github.io/notebook/2021/08/08/clip-vqgan/ to coincide with the growing interest in these models

Keyboard shortcuts

/Search
j kNext / previous blog
EnterOpen selected blog
vVisit the blog site
fFollow / unfollow
lSwitch grid / list view
dSwitch dark mode
g hHome
g pTop posts
g tTopics
g fFollowing
g lLiked
g yHistory
g qFAQ
EscBack / close