ＡＢＡＮＤＯＮＣＵＬＴＵＲＥ

I wrote a post going all the way through how Earley parsing works and how it got there! The final parser is cheap on anything that any other common parsing technology is cheap on. I cover the entire algorithm and all the modernizations. wareya.wordpress.com/2025/09/26/ear…

😵‍💫

🐼×🍨

ひとり遊び中のヨヨ

昨日絵チャで描いてたシグウィン

#ゼンゼロ ちなっちゃん

ナヒーダ🍀🍀🍀

落ちてるﾅﾋ (魔人任務とは関係ありません)

セーラーワンピース風ナヒーダ

フォフォとシッポ

ぐすぐすフォフォ😢

ナヒーダ🍀

久々にやったら幻境おもしろかった

💜🩷

美しい・・・

アイノちゃんとプルーネちゃん🩷💜

#マイナン #ポケモン

new england houses being like "please don't take your shoes off at the front door, i don't want you catching whatever's living on the floorboards"

@SebAaltonen One of the reason 2-color palettes work so well in practice is that the encoding side *loves* them. It's extremely straightforward to find an almost-optimal palette. Finding an almost-optimal 3-color palette might turn into something like gradient descent.

@SebAaltonen It's probably more economical in practice to do something like have an explicit bool-mask for subsets instead of BC7's hardcoded arrangement of possible subsets. You could have two 2-color palettes that share a middle color. The encoder would hate it but it might work.

Has anybody tested texture block/brick compression with 3 color palette? Interpolation between 3 colors is identical to barycentric interpolation on a triangle. You store A+B and calculate C=1-A-B. Seems nicer than the usual 2 color interpolation (BC/ASTC).