Conversation

Notices

1. Embed this notice
   Gianni Rosato (gianni@disobey.net)'s status on Thursday, 10-Jul-2025 02:49:22 JST Gianni Rosato

   Introducing Iris-WebP, a better WebP encoder: https://halide.cx/iris/index.html

   Iris offers up to >20% better compression than libwebp, and trades blows with libaom for AVIF.

   My goal is to build an image-first encoder ecosystem for the modern Web, so that new web image codecs aren't generally superfluous.

   #av1 #webp #compression

   • Embed this notice
     Kai (schweinepriester@mastodon.social)'s status on Thursday, 10-Jul-2025 04:19:27 JST Kai

     in reply to

     @gianni very noice!
     how does it compare to JXL?

   • Embed this notice
     Gianni Rosato (gianni@disobey.net)'s status on Thursday, 10-Jul-2025 04:19:27 JST Gianni Rosato

     in reply to

     • Kai

     @Schweinepriester Here’s a comparison to the latest libjxl release (green) among others. It is hard to compare libjxl sometimes, because it excels on really difficult images but often isn’t as good on average. Also, it is best in the high fidelity range, but we’re measuring from SSIMULACRA2 0-80 for BD-rate (meaning low through high fidelity, but not ultra-high)

   • Embed this notice
     Gianni Rosato (gianni@disobey.net)'s status on Thursday, 10-Jul-2025 04:59:33 JST Gianni Rosato

     in reply to

     • Kornel

     @kornel Thank you! Hoping it will provide value for someone :)

     The chroma subsampling is difficult … I’ve done a couple of things in Iris to mitigate its negative impact, but it remains a challenge.

     As for testing, I think the most valuable test would be the one on the Iris webpage (https://halide.cx/iris/index.html).

     Daala’s subset2 source images are all 4:2:0 which is unfortunate, but the numbers look similar with 4:4:4 subset1 (couldn’t find the same for subset2) favoring AVIF a bit.

   • Embed this notice
     Kornel (kornel@mastodon.social)'s status on Thursday, 10-Jul-2025 04:59:34 JST Kornel

     in reply to

     @gianni Great project! I hoped someone would implement this :)

     A big quality limitation of WebP is forced chroma subsampling. Have you done something in this area specifically?
     Have you controlled for subsampling and luma vs chroma bitrates in your benchmarks?