Skip to content
  • Home Home Home, current page.
  • Moments Moments Moments, current page.

Saved searches

  • Remove
  • In this conversation
    Verified accountProtected Tweets @
Suggested users
  • Verified accountProtected Tweets @
  • Verified accountProtected Tweets @
  • Language: English
    • Bahasa Indonesia
    • Bahasa Melayu
    • Català
    • Čeština
    • Dansk
    • Deutsch
    • English UK
    • Español
    • Filipino
    • Français
    • Hrvatski
    • Italiano
    • Magyar
    • Nederlands
    • Norsk
    • Polski
    • Português
    • Română
    • Slovenčina
    • Suomi
    • Svenska
    • Tiếng Việt
    • Türkçe
    • Ελληνικά
    • Български език
    • Русский
    • Српски
    • Українська мова
    • עִבְרִית
    • العربية
    • فارسی
    • मराठी
    • हिन्दी
    • বাংলা
    • ગુજરાતી
    • தமிழ்
    • ಕನ್ನಡ
    • ภาษาไทย
    • 한국어
    • 日本語
    • 简体中文
    • 繁體中文
  • Have an account? Log in
    Have an account?
    · Forgot password?

    New to Twitter?
    Sign up
keenanisalive's profile
Keenan Crane
Keenan Crane
Keenan Crane
@keenanisalive

Tweets

Keenan Crane

@keenanisalive

Digital Geometer. Michael B. Donohue Associate Professor of Computer Science and Robotics @CarnegieMellon @SCSatCMU, and member of the @GeomCollective.

Pittsburgh, PA
keenan.is/here
Joined January 2008

Tweets

  • © 2022 Twitter
  • About
  • Help Center
  • Terms
  • Privacy policy
  • Cookies
  • Ads info
Dismiss
Previous
Next

Go to a person's profile

Saved searches

  • Remove
  • In this conversation
    Verified accountProtected Tweets @
Suggested users
  • Verified accountProtected Tweets @
  • Verified accountProtected Tweets @

Promote this Tweet

Block

  • Tweet with a location

    You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more

    Your lists

    Create a new list


    Under 100 characters, optional

    Privacy

    Copy link to Tweet

    Embed this Tweet

    Embed this Video

    Add this Tweet to your website by copying the code below. Learn more

    Add this video to your website by copying the code below. Learn more

    Hmm, there was a problem reaching the server.

    By embedding Twitter content in your website or app, you are agreeing to the Twitter Developer Agreement and Developer Policy.

    Preview

    Why you're seeing this ad

    Log in to Twitter

    · Forgot password?
    Don't have an account? Sign up »

    Sign up for Twitter

    Not on Twitter? Sign up, tune into the things you care about, and get updates as they happen.

    Sign up
    Have an account? Log in »

    Two-way (sending and receiving) short codes:

    Country Code For customers of
    United States 40404 (any)
    Canada 21212 (any)
    United Kingdom 86444 Vodafone, Orange, 3, O2
    Brazil 40404 Nextel, TIM
    Haiti 40404 Digicel, Voila
    Ireland 51210 Vodafone, O2
    India 53000 Bharti Airtel, Videocon, Reliance
    Indonesia 89887 AXIS, 3, Telkomsel, Indosat, XL Axiata
    Italy 4880804 Wind
    3424486444 Vodafone
    » See SMS short codes for other countries

    Confirmation

     

    Welcome home!

    This timeline is where you’ll spend most of your time, getting instant updates about what matters to you.

    Tweets not working for you?

    Hover over the profile pic and click the Following button to unfollow any account.

    Say a lot with a little

    When you see a Tweet you love, tap the heart — it lets the person who wrote it know you shared the love.

    Spread the word

    The fastest way to share someone else’s Tweet with your followers is with a Retweet. Tap the icon to send it instantly.

    Join the conversation

    Add your thoughts about any Tweet with a Reply. Find a topic you’re passionate about, and jump right in.

    Learn the latest

    Get instant insight into what people are talking about now.

    Get more of what you love

    Follow more accounts to get instant updates about topics you care about.

    Find what's happening

    See the latest conversations about any topic instantly.

    Never miss a Moment

    Catch up instantly on the best stories happening as they unfold.

    Keenan Crane‏ @keenanisalive 6 May 2020

    Very excited to share #SIGGRAPH2020 paper w/ @rohansawhney1 on "Monte Carlo Geometry Processing" http://www.cs.cmu.edu/~kmcrane/Projects/MonteCarloGeometryProcessing/index.html … We reimagine geometric algorithms without mesh generation or linear solves. Basically "ray tracing for geometry"—and that analogy goes pretty deep (1/n)pic.twitter.com/PQOCtNaY6b

    2:52 PM - 6 May 2020
    • 426 Retweets
    • 1,454 Likes
    • Luna 💙 Thomas Hooper d_gfx Sebastian qwerty Dimon Lers Steffen Christensen Jim (Linxi) Fan Martin Gale
    30 replies 426 retweets 1,454 likes
      1. New conversation
      2. Keenan Crane‏ @keenanisalive 6 May 2020

        Especially for problems with super complicated geometry, not having to mesh the domain provides a massive reduction in real world end-to-end cost. For instance, this model takes 14 hours to mesh and solve w/ FEM, but less than 1 minute to preview with Monte Carlo: (2/n)pic.twitter.com/S7vnhzdXGY

        1 reply 8 retweets 66 likes
        Show this thread
      3. Keenan Crane‏ @keenanisalive 6 May 2020

        As an added bonus, we can do geometry processing directly on nonmanifold meshes, implicit surfaces, instanced geometry, CSG, Bezier curves, NURBS surfaces, etc., without doing any tessellation, sampling, mesh booleans, etc. So how does this all work? Here's the story. (3/n)pic.twitter.com/k5lWl09eZM

        2 replies 3 retweets 44 likes
        Show this thread
      4. Keenan Crane‏ @keenanisalive 6 May 2020

        Once upon a time in rendering, finite element radiosity was the cool new thing. It let you compute some really beautiful lighting effects. But with a catch: you had to nicely mesh your domain. This was a pain, because meshing complex geometry was *hard*—and still is! (4/n)pic.twitter.com/wt8dWD78y2

        2 replies 3 retweets 29 likes
        Show this thread
      5. Keenan Crane‏ @keenanisalive 6 May 2020

        Also, to figure out the illumination, you still had to solve a big global linear system involving every mesh node—even if you only wanted to look at a small region of the scene. (5/n)

        1 reply 1 retweet 19 likes
        Show this thread
      6. Keenan Crane‏ @keenanisalive 6 May 2020

        Monte Carlo rendering takes a different approach: trace rays from the eye, and let them bounce around until they hit a light. Now, instead of meshing, the only geometric query you need is a ray-scene intersection. And you only have to compute the solution where you need it. (6/n)pic.twitter.com/N4rcEkM7M9

        1 reply 2 retweets 37 likes
        Show this thread
      7. Keenan Crane‏ @keenanisalive 6 May 2020

        Though a big impetus for Monte Carlo rendering was more sophisticated illumination, it also enabled extremely challenging *geometry*. Even horrible meshes (from an FEM perspective) still produce gorgeous renderings. As a geometry person, this has always made me jealous! (7/n)pic.twitter.com/MFNtZbzvzE

        1 reply 7 retweets 56 likes
        Show this thread
      8. Keenan Crane‏ @keenanisalive 6 May 2020

        Our paper builds on some terrific but little-used work on stochastic PDE solvers to provide the same capabilities for geometry processing. In short: we replace Kajiya's recursive rendering equation with recursive integral equations for common PDEs (Laplace, Poisson, etc.) (8/n)pic.twitter.com/W758eYqfcA

        1 reply 5 retweets 55 likes
        Show this thread
      9. Keenan Crane‏ @keenanisalive 6 May 2020

        The starting point is Muller's 1956 awesome "Walk on Spheres" algorithm. Say you want to solve a Laplace equation Δu = 0 with fixed (Dirichlet) boundary values g. The mean value property says that u(x₀) equals the average of u over any sphere around x₀. (9/n)pic.twitter.com/ZZszyhfovG

        1 reply 3 retweets 28 likes
        Show this thread
      10. Keenan Crane‏ @keenanisalive 6 May 2020

        To estimate the average, you can sample a random point x₁ on the biggest sphere around x₀ and evaluate u(x₁). In expectation, this gives *exactly* the right value. Of course, you don't know u at x₁. So, recurse. Eventually you can just grab the boundary value g(xₙ). (10/n)pic.twitter.com/hn7qkJRCWk

        2 replies 5 retweets 29 likes
        Show this thread
      11. Keenan Crane‏ @keenanisalive 6 May 2020

        The solution to Δu = 0 is also the expected value where a random walk first hits the boundary (Kakutani's principle). How do you simulate random walks? You could take random steps on a mesh, but then you'd need a mesh! Or you could step in random directions—but how far? (11/n)pic.twitter.com/Z3hRlnzJEV

        1 reply 2 retweets 26 likes
        Show this thread
      12. Keenan Crane‏ @keenanisalive 6 May 2020

        But wait: by symmetry, a random walk starting at some point x₀ is equally likely to exit through any point on a sphere around x₀—no matter how long it takes or what it does inside the sphere. So, to perfectly simulate a continuous random walk, take a walk on spheres! (12/n)

        1 reply 2 retweets 23 likes
        Show this thread
      13. Keenan Crane‏ @keenanisalive 6 May 2020

        Either way, the only thing you need to know about the geometry is the size of the largest empty sphere, which you can get from a closest point query. No meshing required. Otherwise, the code can fit in a tweet—here's a full 3D Laplace solver in C++: (13/n)

        1 reply 1 retweet 24 likes
        Show this thread
      14. Keenan Crane‏ @keenanisalive 6 May 2020

        float solve(Vec3D x0,vector<array<Vec3D,3>> tris,function<float(Vec3D)> g,int N){ float sum=0.f; for(int i=0;i<N;i++){ Vec3D x=x0; float R; do{ R=INF; for(auto t:tris) R=min(R,dist(x,t)); x=x+R*randSphere(); }while(R>1e-3); sum+=g(x); } return sum/nWalks; }

        1 reply 1 retweet 47 likes
        Show this thread
      15. Keenan Crane‏ @keenanisalive 6 May 2020

        The only things missing are (i) how to uniformly sample the sphere and (ii) how to compute point-triangle distance. But even real code is short—here's a 2D Laplace solver (that compiles!) in 100 lines of C++: http://www.cs.cmu.edu/~kmcrane/Projects/MonteCarloGeometryProcessing/WoSLaplace2D.cpp.html … (15/n)pic.twitter.com/sryxNNEiSB

        3 replies 4 retweets 45 likes
        Show this thread
      16. Keenan Crane‏ @keenanisalive 6 May 2020

        In the 2D code, you can toss in any collection of line segments (which can have intersections, holes, etc.) and a function that gives the boundary values—here, a checkerboard (in homage to rendering). It computes solution values at arbitrary points—here, a pixel grid. (16/n)pic.twitter.com/0r5dqqFSsd

        1 reply 3 retweets 33 likes
        Show this thread
      17. Keenan Crane‏ @keenanisalive 6 May 2020

        There's a nice connection here to @_AlecJacobson's generalized winding numbers https://igl.ethz.ch/projects/winding-number/ … which is in essence a grid-free method for a very special PDE—and has enabled super robust mesh booleans, tet meshing, etc. But now we can solve *lots* of PDEs this way! (17/n)pic.twitter.com/l0d4vY0V3s

        1 reply 5 retweets 33 likes
        Show this thread
      18. Keenan Crane‏ @keenanisalive 6 May 2020

        From here you can go crazy, which is what we do in the paper. You can design estimators for other PDEs, do variance reduction, adaptive sampling, etc. And the great thing is that we already have deep knowledge from the rendering community that can turbo-charge this effort. (18/n)pic.twitter.com/FwqwWxNy4L

        2 replies 2 retweets 31 likes
        Show this thread
      19. Keenan Crane‏ @keenanisalive 6 May 2020

        A super cool example, for instance, is that material & light importance sampling strategies from rendering have a direct analogue as Green's function & source term sampling in PDEs. So, we can use multiple importance sampling to reimagine this classic Veach demonstration: (19/n)pic.twitter.com/REusQeZMVS

        1 reply 3 retweets 29 likes
        Show this thread
      20. Keenan Crane‏ @keenanisalive 6 May 2020

        Of course, the whole point is to have fun with geometry. E.g., rather than hard booleans, you can blend geometry together a la @iquilezles. E.g., rather than solve PDEs on a mesh then trace streamlines, we can lazily evaluate points needed for an ODE integrator. And so on. (20/n)pic.twitter.com/kQy03EdqVY

        1 reply 5 retweets 43 likes
        Show this thread
      21. Keenan Crane‏ @keenanisalive 6 May 2020

        Geometric algorithms built on top of this framework share common benefits: parallel implementation is trivial; the main cost (as in ray tracing) is just doing BVH queries. Like ray tracing, we can also focus computational effort on just a region or slice plane of interest: (21/n)pic.twitter.com/YyIzHGa2dV

        2 replies 3 retweets 35 likes
        Show this thread
      22. Keenan Crane‏ @keenanisalive 6 May 2020

        Finally, this all fits into existing geometry processing pipelines. Need a solution value at a vertex? Run our black-box solver. …Ok, if I say more I'll probably just rewrite the paper here on Twitter! As you can tell, we're really excited about the possibilities. :-) (n/n)

        9 replies 3 retweets 86 likes
        Show this thread
      23. End of conversation

    Loading seems to be taking a while.

    Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.

      Promoted Tweet

      false

      • © 2022 Twitter
      • About
      • Help Center
      • Terms
      • Privacy policy
      • Cookies
      • Ads info