Here's my 3DV talk, in chapters: 1) Intro / NeRF boilerplate. 2) Recent reconstruction work. 3) Recent generative work. 4) Radiance fields as a field. 5) Why generative video has bitter-lessoned 3D. 6) Why generative video hasn't bitter-lessoned 3D. 5 & 6 are my favorites.

Introducing World Tracing Generative pixel-aligned geometry, beyond the visible. Faithful to your pixels. Complete in 3D. One image in — objects, scenes, even dynamic worlds emerge in full geometry, every point traced back to the pixel it came from. 🌐 haoz19.github.io/world-traci… 🧵 [1/6]

Up until now, the models had been using primitives for building objects. I couldn’t believe the Boeing 747 shared by @victormustar was also built from primitives, so I investigated. With Three.js DevTools I could see the geometry was fully custom. The source was minified, so I downloaded it and asked Claude to analyze it. huggingface.co/spaces/victor… Turns out these models have now learned to generate their own custom geometry constructors when the library doesn’t provide them.

There's an interesting procedural generation -> generative AI -> procedural generation ouroboros closing in on itself right now. GenAI is basically data driven procedural generation, and now the LLMs are leveraging perlin noise height maps and fractal tree models.

昨日作ったゲームのデザインガイドを元に、Claude Fable 5でローポリ3Dモデル量産してみた！ 初稿が可愛すぎて……ほぼ手直しゼロ😂Claudeすごい！ ローポリ3Dモデルを生成できるプラグイン作ってみたい！！ #claude #Lowpoly #3Dモデル

We teamed up with researchers at Meta @RealityLabs to stress-test what's actually possible with Gaussian Splatting, and especially with volumetric video. Reconstructing buildings on a sunny day or celebrities in controlled studio conditions has gained a lot of traction. But what if the goal is to bring the viewer to an actual electronic rave, with zero compromises allowed? High-frequency strobing lights, fog machines filling the space with haze, complete darkness that clips the camera sensors to black. It was brutal, but together we pulled it off. We're proud to present a teaser of our joint project: an actual performance of 'Patience' by NYC electronic duo @membamusic, captured in full volumetric format and optimized for a wide range of devices. We didn't design this reconstruction around what the tech allows. Instead, we customised our pipeline to make the experience as immersive as possible. More details coming soon.

Good news everyone, kernel methods are AGI, pack it up

Hey @NeurIPSConf, your "strict policy prohibiting reviewers and ACs from using LLMs as part of the reviewing process" --- what exactly does this mean? Can I use google search to learn about math while reviewing, or must I avert my eyes from the AI summary?

What if my research is too bullshit for Claude to even care about rerouting it😔

Millions of Gaussian splats, rendered directly in the browser. For ICARE, the creative team designed and iterated 7 worlds with Marble, then streamed them in real time using Spark 2.0 and Three.js Here's how the pipeline works ↓🧵

Explore worlds inspired by Da Vinci, Tesla, Hokusai, Frida Kahlo, and other visionaries from history. Created with Marble. Millions of Gaussian splats, streamed directly in your browser (desktop only). Play Icare: icare-game.world

Apple just released 3D gaussian splatting on Apple Maps 🤯 This is gonna hit the Urbanist community the way crack cocaine hit 1980s New York.

ugh, you do "compiler coding"? I'm a serious programmer so I still do *actual* coding (plugging cables into a house-sized ENIAC)

If your conference talk is good, you should upload it somewhere (I use YouTube). If your talk is not good enough for you to feel comfortable uploading it somewhere, then it is *definitely* not good enough to present at the conference, and you should fix it.

I want to offer some unsolicited advice to computer vision researchers jumping into robotics. Don't focus too much on VLMs, VLAs etc. That's fine, but the real action is at the sensorimotor level. Most of the open problems in robotics are in manipulation, which is about hand-object interaction, and contacts and forces are central. Proprioception and tactile sensing are as important as vision. Don't get seduced by cherry-picked demos. You can't do robotics without doing robotics.

Introducing Magenta RealTime 2 (MRT2): the live music model you can play as an instrument. MRT2 offers MIDI and prompt controls, and runs natively on a MacBook with <200ms latency. Open weights. Open source inference engine. Suite of apps and plugins. Hear what it can do and try it out for yourself below 🧵

My advice to PhD students in 2026: 1) If your advisor hasn't logged >100 hours in a modern agentic IDE, stop listening to your advisor. 2) Write your next paper *inside your codebase* as a .tex (with 80 char word-wrap). Force your advisor to read and make edits inside the IDE.

Introducing Ideogram 4.0: the best open image model in the world. Think it. Make it. Own it. Download the weights, fine-tune on your own data, and run it on your hardware. Live on every Ideogram plan and the API today.

I'll be giving a "exactly how bitter lesson'ed is all of 3D computer vision?" talk at the Bitter Lessons CVPR workshop tomorrow at 10:30am, Room 3A-3D. Here's a slide with the overall thesis.