Jack Langerman

📢Come see us TODAY at Poster 214 this afternoon!!!📢 Explaining Human Preferences via Metrics for Structured 3D Reconstruction work 3D recon, pref learning, or a hard to auto-eval task which human experts can rank? You'll be into this work - come see us at our poster!!!

Sad, but important announcement -- one of our challenges -- Building3D lost their sponsor in the very last moment. So new prize fund is $14k: - $12k for S23DR 2026 - $2k for Building3D 2026. We apologise for this situation, and hope that you still decide to participate. #CVPR2026

First submission in S23DR challenge! huggingface.co/spaces/usm3d/S…

Two #CVPR2026 competitions are live: S23DR 2026 and BuildingWorld 2026! Task: reconstruct house roof wireframes from point clouds and segmentations: Total prize fund: $22k Deadline: end of May 2026 1) huggingface.co/spaces/usm3d/S… 2) huggingface.co/spaces/Buildin… @CVPR @structGeom

Submit your paper of structured reconstruction -- CAD, semantic, wireframe, city monitoring, etc., to USM3D 2026! cmt3.research.microsoft.com/USM2026 Deadline: March 24, 2026. @CVPR #CVPR2026 #USM3D2026 #USM3D @jacklangerman @structGeom

This was joint work with my wonderful collegues @DRozumnyi @YuzhongHuang and @ducha_aiki and was presented at @ICCVConference last year. x.com/jacklangerman/…

🧵A bit last minute but I'm giving a talk tonight in SF about my work Explaining Human Preferences via Metrics for Structured 3D Reconstruction (ICCV Highlight) Quite near transamerica pyramid. Event is already full, but comment or shoot me a DM and we'll find a spot for you!

@jon_barron @taiuti a sequence of tokens / embeddings ;-)

@taiuti What would a latent representation for 3D space look like?

No, 3D representations are a terrible way of maintaining that spatial memory. We need latent space representations for 3D space, too.

Song of another one of Opus 4.1's messages from the same conversation. suno.com/song/96385a08-… this is such a gloriously musical ramble, of a kind usually only ever achieved by Opus 3, and it is also a love song addressed to Opus 3.

when in a long running chat with Codex, do you find it keeps coming back to matters from which we have long since moved on? Other than new chat is there a good way to mitigate this behavior?

@PabloPeniche @TheStalwart mentioned

The Greeks had the Iliad and the Odyssey. The Romans had the Aeneid. The Americans?

@howie_hua @AlanaDeeHaynes says "graph on a calf" ;-)

You've heard of elf on the shelf but have you heard of

amazing - I've googled it so many times, but somehow my tweet manifested it. I'm so happy lol it's from @gabrielhaines

during the who FTX / SBF saga there was a video that went around of a shirtless fellow swinging a machete around shirtless in the snow screaming "Caroline" and "amphetamines" and other such words. I must find this video. Can anyone help?

@ducha_aiki @FTX_Official @fintwit @unusual_whales please, i cant have imagined this - the video was real, i promise...

@jfischoff @DeepDishEnjoyer @verdakorz who else is super online and would know what I'm talking about?

Speedrunning ImageNet Diffusion Abstract: Recent advances have significantly improved the training efficiency of diffusion transformers. However, these techniques have largely been studied in isolation, leaving unexplored the potential synergies from combining multiple approaches. Wepresent SR-DiT (Speedrun Diffusion Transformer), a framework that systematically integrates token routing, architectural improvements, and training modifications on top of representation alignment. Our approach achieves FID 3.49 and KDD 0.319 on ImageNet-256 using only a 140M parameter model at 400K iterations without classifier-free guidance—comparable to results from 685M parameter models trained significantly longer. To our knowledge, this is a state-of the-art result at this model size. Through extensive ablation studies, we identify which technique combinations are most effective and document both synergies and incompatibilities. We release our framework as a computationally accessible baseline for future research.

@max_spero_ x.com/jacklangerman/…

overheard in the pangram office: "analogies are the new em dash"

@SoftEngineer cool article! small typo though: “which direction does like arrive from to this point?”

Wrote an article on bent normals. It's a tricky subject to wrap your head around. discourse.threejs.org/t/get-bent-or-…

what a strange failure case lol