Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

Reply to @sun@shitposter.world
rees@rees@pleroma.ree.social (2026-10-06 06:58:48)
@sun are you using codebooks/spherical quantization
---Reply---
Blurry Moon@sun@shitposter.world (2026-10-06 07:18:22)
@rees I'm gathering all the colors then grouping them by a threshold using oklab colorspace value then histogramming them. then refining the process not sure yet how

---Replies---

rees@rees@pleroma.ree.social (2026-10-06 07:25:57)
@sun >then refining the process not sure yet how
you're doing perceptual coding, which you can improve using learned or fixed codebooks. you could have claude code spawn 18 agents to figure out the most efficient vector quantization. I did something similar for a better alternative to thumbhash, although I haven't put it up on a repo yet
---Attachments---
image: https://pleroma.ree.social/media/af/3b/75/af3b75e3ec5883dc8d6014247453735e88a85831dd584942d6c653c8e42c2c2d.webp