A much needed step in world modeling. By generating 360° panos, world models will truly track everything happening around them, serving as a better form of memory. Huge shoutout to @Dazitu_616 for spearheding this awesome work. It was amazing to work with you!

Adam deserves the award, but in Singapore everyone still uses SGD

All results are unedited outputs from a single diffusion model. There is no NeRF or any postprocessing steps involved. Please check out our website for many more samples! https://t.co/yojdRhgNEn

While still nowhere near perfect, 4DiM excels at very difficult cases like 360º from a single image:

Or create videos in between images (without any camera poses):

4DiM can also manipulate videos, e.g. we can re-create them with different cameras:

Scenes are difficult because there is little training data with camera poses, and poses are usually noisy / lack meaningful scale. To get around this, we leverage large-scale (unposed) video data. This lets us also control time:

[[THREAD]] Happy to announce 4DiM, our diffusion model for novel view synthesis of scenes! 4DiM allows camera+time control with as few as one input image. Joint work with @srbhsxn* @lala_yi_li* @taiyasaki @fleet_dj *equal contribution

Excited to share latest ✨Med-Gemini✨ additions - our new research unlocks possibilities in medical data analysis with 3 new models built upon Gemini 1.5 that can handle 2D medical images, and for the first time genomic risk score & 3D radiology scans. https://t.co/QKz0NZqpDH

Excited to share our work on Video Interpolation with Diffusion Models. https://t.co/Ckt7MtiTnf VIDIM generates plausible short videos given a start and end frame. Joint work with @watson_nn, @erictabellion , @holynski_ , @poolio and @jannekontkanen

Very excited to share VIDIM, our diffusion model for video interpolation. Cheap and high quality video diffusion will become a reality. I have also believed for a while now that using pixels as inputs is severely underrated-- there is so much more signal in there v.s. text.

Fast sampling with 'Multistep Consistency Models': We get 1.6 FID on Imagenet64 in 4 steps and scale text-to-image models, generating 256x256 images with 16 steps. Guess which row is distilled? With @emiel_hoogeboom @TimSalimans Arxiv: https://t.co/BH7HzIGsgI

# on shortification of "learning" There are a lot of videos on YouTube/TikTok etc. that give the appearance of education, but if you look closely they are really just entertainment. This is very convenient for everyone involved : the people watching enjoy thinking they are

Hiring Research Scientists within @GoogleDeepMind - Toronto to join our team & advance the next generation of medical AI, develop cutting-edge LLMs & Multi-modal models to tackle real-world healthcare challenges. Please submit your interest through: https://t.co/FJBs3h7Nvr

Las palabras se las lleva el viento, Presidente En educación: seguimos fracasando las pruebas PISA Transparencia: No hay mejoras desde el 2019 (Transparencia Internacional) Pobreza: incrementa desde 2019 (Banco Mundial) Homicidio: incrementa desde 2019 (Procuraduría)

We have a student researcher opportunity in our team @GoogleDeepMind in Toronto 🍁 If you’re excited about research on diffusion models, and generative video models, please fill the form : https://t.co/3svxJfm8nO and apply here: https://t.co/82FhJvhV4B

Excited to share ReconFusion! 3D reconstruction of real-world scenes from only a few photos, powered by diffusion priors: https://t.co/BjOBi7bIth w/ amazing team @ChrisWu6080 @BenMildenhall @philipphenzler @KeunhongP @RuiqiGao @watson_nn @_pratul_ @dorverbin @jon_barron @poolio

Logging into Twitter after the CVPR deadline...

El Procurador General de la Administración ha concluido SIETE violaciones a la Constitución Política de la República de Panamá en la Ley 406 (contrato minero) y cierra magistralmente su escrito con esta cita del ilustre Carlos Iván Zúñiga: "hay momentos supremos y de conciencia