The 3D Gaussian Splatting paper has (rightfully!) stirred up a lot of attention and some friends asked me for a technical take on it, so here we go:
TLDR: Fast, simple, outstanding level of detail, with some caveats (see below).
1/x
3D Gaussian Splatting for Real-Time Radiance Field Rendering
paper page: https://t.co/q9P2DmfJyP
Radiance Field methods have recently revolutionized novel-view synthesis of scenes captured with multiple photos or videos. However, achieving high visual quality still requires neural networks that are costly to train and render, while recent faster methods inevitably trade off speed for quality. For unbounded and complete scenes (rather than isolated objects) and 1080p resolution rendering, no current method can achieve real-time display rates. We introduce three key elements that allow us to achieve state-of-the-art visual quality while maintaining competitive training times and importantly allow high-quality real-time (>= 30 fps) novel-view synthesis at 1080p resolution. First, starting from sparse points produced during camera calibration, we represent the scene with 3D Gaussians that preserve desirable properties of continuous volumetric radiance fields for scene optimization while avoiding unnecessary computation in empty space; Second, we perform interleaved optimization/density control of the 3D Gaussians, notably optimizing anisotropic covariance to achieve an accurate representation of the scene; Third, we develop a fast visibility-aware rendering algorithm that supports anisotropic splatting and both accelerates training and allows realtime rendering. We demonstrate state-of-the-art visual quality and real-time rendering on several established datasets.
@guidoquelle Ich halte es gesellschaftlich nicht nur für akzeptabel, sondern auch für notwendig, Personen in Machtpositionen, die nachweislich gelogen haben, auch als Lügner zu bezeichnen.
Our second round was rough. I think I pitched every VC on the planet. Some of the weirder ones:
- I arrive at a Chinese VC firm on Sand Hill Road. The building is locked, lights off, nobody inside. I wait 10 minutes after meeting time. A car pulls up. The partner emerges, she has a large pack of toilet paper under one arm. She unlocks the building, turns on the lights, leads me inside. She disappears into the toilets with her rolls of paper. Then, toilets attended to, we start the pitch.
- Offices of a small deep tech investor. Receptionist greets me and leads me to the meeting room. I set up, wait, wait, wait. Eventually someone walks in, invites me to start. Something feels a bit off. It gradually dawns on me that I am pitching the receptionist. The partner had double-booked herself and apparently decided this was preferable to rescheduling.
- Pitching a London VC. He tells me he likes me as a founder but my business will never work. I've never had someone say this before, usually VCs like to preserve optionality. But he's very explicit. "You might close this round, but you will just waste a year or two of your life and then fail, this idea can't work"
A month later, I get a message from him. "I hear (big name) invested, can we get into your round?". Dude, seriously?
- Another London VC. I'm chatting to the associate while we're waiting for the partner to arrive. He tells me about one of their portfolio cos that needs a bridge round, he's working on squeezing them as much as possible. He's quite direct about it. Eh.. thanks for the transparency I guess, you seem like a great firm.
- The all time weirdest. A French VC with a London office. The associate meets me, as we're going up to the meeting he says something about how the partner can be a little unconventional. We start the pitch.
"What did your father do?" the partner asks me in a thick French accent. OK, this is a new one. I say he trained as a theoretical physicist, but then went into business. "Aha! Your father was a failure!"
"What did your mother do?" I say she was a biochemist and then became a school teacher. "Also a failure!" he exclaims. The associate is visibly dying inside. Is this some strange test, or is he just an asshole? I have a hundred employees and we need funding. "Would you like to hear about my company?" I ask him.
La paradoja de la transición energética está aquí.
¿Limpiar paneles usando el sol? Parece obvio, pero la IA en desiertos chinos ya hace autónoma esta industria.
El polvo reduce un 30% la eficiencia solar. Sin agua, este robot mitiga pérdidas millonarias a coste cero.
¿Sustituirá la automatización total al mantenimiento humano en Europa?
@bubbleboi TPUs fight the same fight against NV as AMD, Intel and everyone else.
Virtually 100% of the cutting edge ML research these days targets CUDA and unless you give these researchers $$$ to make it perform on alt chips it simply won't happen.
@bubbleboi As a separate point why wouldn't this cycle be larger than even the 90s? This is the first cycle where compute itself, not consumer demand, begets more compute demand
Yes, for sure.
At the end of the day there's always foundational research of the core math and then there's lots and lots of research on top to make things more efficient / scalable / flexible etc.
What I'm saying is: the core is mature, but most things on top still have a lot more room to run.
Du konntest keines meiner Argumente widerlegen, dafür postest du noch eine Grafik die China's AKW Expansion zeigt 👍
@grok bitte kläre @DerClue auf bzgl. Bodenschäden, Mikroplastik/Materialabrieb, Flächenversiegelung, Waldrodung, Verdrängung von Wildtieren, Lärm und Vogelkollisionen bei Windkraft.
Von welchen Alternativen sprichst du?
Solar haben wir bereits Überkapazität, trotzdem müssen wir teuer den Atomstrom der Franzosen kaufen.
Hydro gibt's hier kaum. Wind ist wie Solar, nur 10x dreckiger. Biogas etc. nicht skalierbar.
Man kann sich Atomkraft beliebig schlecht reden und 1000 Argumente erfinden. Am Ende des Tages siegt dann aber doch die Physik.