Thomas Stevenson

VFX丨Rigging丨Pipeline | Computer Graphics丨Animation & Rigging Software@Netflix Animation Studios | Previously Unity, Wētā, MPC, Aardman | 🏴󠁧󠁢󠁷󠁬󠁳󠁿🇬🇧🇳🇿

[Francis Stokes @fstokesman]

This is a thread about the CORDIC algorithm, and why it's been living rent-free in my brain for the last couple of weeks 🧵

[Thomas Stevenson @TomStevensonTD]

@Muream That’s cool

[Peyman Milanfar @docmilanfar]

The perpetually undervalued least-squares:

minₓ‖y−Ax‖²

can teach a lot about some complex ideas in modern machine learning including overfitting & double-descent.

Let's assume A is n-by-p. So we have n data points and p parameters

1/9

[Peyman Milanfar @docmilanfar]

Almost every technical person knows about least-squares(LS) but most don’t know about *total* least-squares(TLS).

These measure fitting error differently: LS minimizes sum of sq. vertical distances whereas TLS minimizes the sum of orthogonal distances from data to fit line

1/2

[Thomas Stevenson @TomStevensonTD]

New open source project from our team at Animal!

[Animal Logic @animallogic]

As part of our ongoing commitment to the open-source community, we are excited to announce the open-source release of OMX, a user-friendly and performant extension to @Autodesk #Maya Python API. 🔗 https://t.co/CcvZcmnpVQ
#AnimalLogic #SIGGRAPHAsia #opensource

[Thomas Stevenson @TomStevensonTD]

@fulagann @FreyaHolmer It’s in Real Time Collision Detection. It’s the basis for doing closest point on line. I think it doesn’t have a name because it tends to be part of other more useful functions geometric functions.

[Thomas Stevenson @TomStevensonTD]

@FreyaHolmer Long threads get hard to navigate, especially for finding other peoples replies
This has inspired me to implement something similar in Maya and I've already had to search hard for a specific reply a few times lol
Great work so far btw

[The Video Game Library 📚 ➡️ Over on Bluesky @TheVGLibrary]

Alright #GameDev friends - finally a post for you today!

Admittedly a little older, the "GAME PROGRAMMING GEMS" collection offered THOUSANDS of pages of reference for aspiring #VideoGame developers.

Now all catalogued into #TheVideoGameLibrary:

👉 https://t.co/LPvQcQrJ8T

[Sebastian Aaltonen @ SIGGRAPH @SebAaltonen]

We just changed most of our 4x4 matrices to 4x3 matrices (named Matrix4 and Matrix4Affine currently in our code base).

Matrix4 * Matrix4 and Matrix4 * MatrixAffine return Matrix4. But MatrixAffine * MatrixAffine returns MatrixAffine. Matrix4 can't become affine...

[Harry Houghton @Haz_Houghton]

Procedurally splitting a blendShape target using curve parameters and De Boor's algorithm. The result is normalized targets which match the original target and have smooth falloffs, very similar to Pixar's "3 Curve Principal"

[Thomas Stevenson @TomStevensonTD]

@Haz_Houghton Gottcha, thanks for sharing! Nice results.

[Animal Logic @animallogic]

And that's a wrap on @siggraph and @digiproconf for another year! 🤝Our team had a blast sharing what we've been working on and learning more about the industry we love. Great to see so many new and familiar faces over the past week!
#AnimalLogic #SIGGRAPH #DigiPro

[Freya Holmér @FreyaHolmer]

quaternionic hermite interpolation w. arbitrary axis start/end angular velocities. it's so beautiful ;-;

[Thomas Stevenson @TomStevensonTD]

@FreyaHolmer Yep, of course! I was working through it today and it makes perfect sense now.
This is a neat solution. I've used hermite interpolation on top of slerp to get smoother passes through normal samples, but I never thought of this. Super useful in some cases.

[Sebastian Aaltonen @ SIGGRAPH @SebAaltonen]

Let’s talk about DoD, or data-oriented design.

There’s a lot of misconceptions what DoD is and how to apply it to your architecture design.

DoD is not a data structure or algorithm. And it is certainly not equivalent to ECS with batched inner join queries.

Thread…

[Dr Andrew Digby @takapodigs]

Tonight there are #kakapo in the forest of the North Island of Aotearoa for the first time in over a century. We released four into Maungatautari Sanctuary Mountain today. #conservation #returnofthekakapo

[Ben Grimmer @prof_grimmer]

I've proven the strangest result of my career..
The classic idea that gradient descent's rate is best with constant stepsizes 1/L is wrong. The idea that we need stepsizes in (0,2/L) for convergence is wrong.
Periodic long steps are better, provably.
https://t.co/Yqe5PUKgxI