https://xitter.cf/t.co/hynShC9Qrc
Joined May 2009
- Tweets102
- Following170
- Followers249
- Likes97
Dan Rosenbaum retweeted
Cat's out of the bag (and the cat is actually a unicorn 🦄)!
Come do cool research with us in sunny Barcelona 🌞fundamental.tech/research
Ever wondered how underwater scenes would look like without water?🪸🤿
Come check our work at #ECCV24 in Milan tomorrow.
Osmosis: RGBD Diffusion Prior for Underwater Image Restoration.
Opher Bar Nathan, Deborah Levy, @TreibitzL.
Poster#237 16:30
Code: osmosis-diffusion.github.io/
Dan Rosenbaum retweeted
youtube.com/watch?v=7rTFuVrW…
Sea Thru #NeRFs enable us to model the underwater environment and as a result remove the water and expose how the scene looks like without water. Come see our poster (number 6) Tue AM in #CVPR2023.
@dakkaynak
@SimonKorman
@danrsm
Dan Rosenbaum retweeted
Come see our poster (number 6)- we use a novel underwater #nerf formulation to remove the water and restore the scenes #CVPR2023 @dakkaynak @SimonKorman @danrsm – at Vancouver, British Columbia
NeRF goes underwater! 🤿
How do you synthesize novel views underwater? Happy to share results from our new #CVPR2023 paper: "SeaThru-NeRF: Neural Radiance Fields in Scattering Media", Deborah Levy, Amit Peleg , Naama Pearl, Dan Roesnbaum, Derya Akkaynak, Simon Korman, Tali Treibitz.
@dakkaynak @SimonKorman @danrsm @CVPR #NeRFs #nerf #3dmodeling #Photogrammetry #ocean #oceans
youtu.be/oRMvTBBARKE
Scaling up functa
Previously we had introduced *functa*, a framework for representing data as neural functions (aka neural fields, INRs) and doing deep learning on them.
In our recent work *spatial functa* we show how to scale up the approach to ImageNet-1k 256x256.
📝arxiv.org/abs/2302.03130
Dan Rosenbaum retweeted
Functa is now open-sourced!💻 Hope it's useful to those of us exploring INRs/Neural fields 😀
github.com/deepmind/functa
Ever wondered why deep learning is always done on array data?🤔 Happy to announce our work:
From data to functa: Your data point is a function and you can treat it like one
📝arxiv.org/abs/2201.12204 w/ @emidup @arkitus @DaniloJRezende @danrsm, to appear in ICML22
If you're at #ICML2022 come see the talk and poster today of our work on treating data points as functions
w/ @emidup @hyunjik11 @arkitus @DaniloJRezende
Ever wondered why deep learning is always done on array data?🤔 Happy to announce our work:
From data to functa: Your data point is a function and you can treat it like one
📝arxiv.org/abs/2201.12204 w/ @emidup @arkitus @DaniloJRezende @danrsm, to appear in ICML22
Check out our new 💃functa🕺work!
How can function representation lead to psychedelic cars?
w/@hyunjik11 @emidup @arkitus @DaniloJRezende
Ever wondered why deep learning is always done on array data?🤔 Happy to announce our work:
From data to functa: Your data point is a function and you can treat it like one
📝arxiv.org/abs/2201.12204 w/ @emidup @arkitus @DaniloJRezende @danrsm, to appear in ICML22
If you're interested in dynamic protein structure, cryo-EM data, and/or inverse graphics approaches, come see our talk and poster today at the MLSB workshop at NeurIPS 3:30PM GMT! mlsb.io/
Proteins are not static bricks! Feasibility study to infer a continuous distribution of all states using an end-to-end model from Cryo-EM images to atom coordinates: arxiv.org/abs/2106.14108.
@danrsm, @GarneloMarta, @MichaelZielins, @JonasAAdler, @arkitus, @CarlDoersch, @pushmeet
Dan Rosenbaum retweeted
@fraser_lab and I, along with many colleagues (including @irisdyoung), held a journal club on the exciting new @DeepMind paper 'Inferring a Continuous Distribution of Atom Coordinates from Cryo-EM Images using VAEs' arxiv.org/abs/2106.14108.
Dan Rosenbaum retweeted
Heartbroken to learn of the untimely passing of Tali Tishby. A profound thinker and a gentle and fine human: a great loss. en.m.wikipedia.org/wiki/Naft…
Dan Rosenbaum retweeted
Excited to share "Volume Rendering of Neural Implicit Surfaces" (VolSDF): a volume rendering framework for implicit neural surfaces, allowing to learn high fidelity geometry from a sparse set of input images.
with @thoma_gu @yoni_kasten @lipmanya
arxiv.org/abs/2106.12052
(1/8)
Dan Rosenbaum retweeted
Cryo-EM: fascinating open vision problem I wouldn't even know about without this collab. Images are so noisy that each one gives only a little info on the 3D shape. We need Bayesian inference in generative models to explain all images w/ a physically-plausible state distribution.
Proteins are not static bricks! Feasibility study to infer a continuous distribution of all states using an end-to-end model from Cryo-EM images to atom coordinates: arxiv.org/abs/2106.14108.
@danrsm, @GarneloMarta, @MichaelZielins, @JonasAAdler, @arkitus, @CarlDoersch, @pushmeet
Dan Rosenbaum retweeted
Infer a continuous distribution over atom coordinates from cryo-EM images using a VAE. The VAE decoder includes a model of how atoms generate images!
@danrsm @GarneloMarta @MichaelZielins @JonasAAdler @arkitus @CarlDoersch @ORonneberger
@pushmeet
arxiv.org/abs/2106.14108.
Dan Rosenbaum retweeted
Interested in generative models, 3D computer vision or inverse graphics?
We use ideas and techniques from these fields to show the possibility of imaging very small objects (e.g. proteins) more effectively.
In this setting we cannot fall back to supervised learning!
Proteins are not static bricks! Feasibility study to infer a continuous distribution of all states using an end-to-end model from Cryo-EM images to atom coordinates: arxiv.org/abs/2106.14108.
@danrsm, @GarneloMarta, @MichaelZielins, @JonasAAdler, @arkitus, @CarlDoersch, @pushmeet
Modelling protein structure dynamics with inverse graphics: from cryo-EM images to a distribution of atom coordinates arxiv.org/abs/2106.14108
@JonasAAdler, @ORonneberger, @GarneloMarta, @MichaelZielins, @arkitus, @CarlDoersch, @pushmeet
Dan Rosenbaum retweeted
Proteins are not static bricks! Feasibility study to infer a continuous distribution of all states using an end-to-end model from Cryo-EM images to atom coordinates: arxiv.org/abs/2106.14108.
@danrsm, @GarneloMarta, @MichaelZielins, @JonasAAdler, @arkitus, @CarlDoersch, @pushmeet
Dan Rosenbaum retweeted
By framing generative modeling as learning distributions of functions, we build models that act entirely on continuous spaces, independently of data resolution 🌿
📄 Paper: arxiv.org/abs/2102.04776
💻 Code: git.io/JG2KV
with @yeewhye @ArnaudDoucet1