Haochen Wang

FastMap: Revisiting Dense and Scalable Structure from Motion Jiahao Li, @__whc__, @mzubairirshad, @vslevic, Matthew R. Walter, Vitor Campagnolo Guizilini, @gregshakh tl;dr: replace BA with epipolar error+IRLS; fully PyTorch implementation arxiv.org/abs/2505.04612

@emiel_hoogeboom @tejmensink @JonathanHeek @KayLamerigts @RuiqiGao @TimSalimans Does posthoc EMA matter for high-res pixel space diffusion?

Is pixel diffusion passé? In 'Simpler Diffusion' (arxiv.org/abs/2410.19324) , we achieve 1.5 FID on ImageNet512, and SOTA on 128x128 and 256x256. We ablated out a lot of complexity, making it truly 'simpler'. w/ @tejmensink @JonathanHeek @KayLamerigts @RuiqiGao @TimSalimans

@_onionesque There are definitely ppl at UChicago/TTIC who would sign up for a mini course if you offer them. Quite a few of us work on 3D here.

@__whc__ Hey Haochen! Nice to see you here! That was an embarrassing 'typo' haha. I meant guinea pigs. Because that is how I can learn, but I don't get the chance! twitter.com/_onionesque/st…

I have taught using this book twice. Once to a study group in 2019 at Brown, and once late last year to a group of undergrads over Saturdays. Looking for a fresh set of scapegoats to also finish the last part. I always recommend this above the Langs and the Serres for beginners.

@lipmanya @sedielem @RickyTQChen @helibenhamu @mnick @lematt1991 for the vector field, the regression target in eq15 is some coef * epsilon + some coef * x_0, it is similar to the precondition discussed in Tero EDM work~(eq8)

@magifactory @_akhaliq There’s a GitHub issue on this. Let’s move the discussion there. I am not supposed to talk on social media while the work is under peer review. I will try some simple method this week and update under that issue.

"A high quality photo of a delicious burger"

@uguryektabasak Sry about the low res img and depth map. We will add the code for sub pixel rendering shortly. It slows things down but for a final vis round like the teaser it is good.

#AI #GenerativeAI #text2art

[Text-to-3D]: A thread 🧵 Today, researchers at @TTIC_Connect and @LifeAtPurdue published a new paper. pals.ttic.edu/p/score-jacobi… This magical work makes #NeRF scene generation from text possible.

@KL_Div @PengShichong Is conditional VAE a potential baseline to compare against?

Excited to share our latest work, which generates high-res images non-adversarially without the use of a discriminator. Benefits include: (1) no mode collapse and (2) no training instability. Joint work with @PengShichong Paper and Code: niopeng.github.io/HyperRIM/ Thread 1/n

@sunniesuhyoung Looks like some sorts of e-sports

I don't know how/why South Korea's broadcasting companies put so much effort into election graphics, but these are pretty crazy and awesome🤯 twitter.com/i/events/12506…