Introducing DiffusionLight---a simple yet effective technique to estimate lighting from any in-the-wild input image. How? ... by inpainting a chrome ball into the image with diffusion models! (1/3) paper: arxiv.org/abs/2312.09168 diffusionlight.github.io huggingface.co/DiffusionLight…

This thing is mind-blowing. I'm literally shaking now

“Occasionally, in times of worry, I’ve longed to be stylish, but on second thought I say no—just let me be myself—and express rough, yet true things with rough workmanship.”⁣⁣ — Vincent van Gogh In times of doubt, listen to #VanGogh—born March 30, 1853—and just be yourself. ❤️

Amazing talk from Lenore and Manuel on AI consciousness. Their report "AI Consciousness is Inevitable: A Theoretical Computer Science Perspective" is also a delight to read: arxiv.org/abs/2403.17101

From Dumas to AI: 𝓢𝓲𝓰𝓷𝓜𝓾𝓼𝓴𝓮𝓽𝓮𝓮𝓻𝓼 unites face, hands, and pose features in a quest for accurate sign language translation. Learn how we are slashing compute needs by 97% in our new study: arxiv.org/abs/2406.06907 #DeepLearning #AIForAll

Just landed in the beautiful Seattle! Looking forward to presenting at #cvpr2024 and making some new friends

Thrilled to share our latest research, Intrinsic-LoRA, at the #CVPR2026 Workshops in the vibrant city of Seattle! The atmosphere here was buzzing with so much exciting work and it's incredible to connect with so many talented individuals in the field.

It's happening today in the afternoon at Brown 1, MiCo Milano! We have invited some of the brightest minds in the field to talk about what and how do generative models learn from the visual world.

Thanks to our amazing speakers (David Forsyth, David Bau, Shalini De Mello, Yossi Gandelsman, Phillip Isola, and Ziwei Liu) and our spotlight speakers (Anton Obukhov, Daniel Winter, Samyadeep Basu, Lorenzo Olearo, Anisha Jain, Xindi Wu, and James Burgess (at CVPR)) for their great talks and for

👀👀👀

Frozen in Motion — Still Life Made with #Ray2 Camera Motion Concepts #aivideo #surrealism #cinematography

Why does Adam outperform SGD in LLMs training? Adaptive step sizes alone don't fully explain this, as Adam also surpasses adaptive SGD. Is coordinate-wise adaptivity the secret? Not entirely—Adam actually struggles in the rotated parameter space! 🧵 (1/6) arxiv.org/abs/2410.08198

TTIC is located on the University of Chicago campus, and students receive full access to University of Chicago courses, campus facilities, and resources. Learn more and apply to our PhD program in #computerscience by Dec. 9 (NO FEE): buff.ly/Qw8njWB