GenWildSplat
Feedforward 3D models are awesome, but they can't handle in-the-wild images with varying illumination.
Check out GenWildSplat that fills this gap! https://genwildsplat.github.io/
CVPR 2026 was a blast! 馃コ
It was great meeting old and new friends and presenting our work.
summarized below if you missed it! 馃У
Edit by Track
Video editing requires precise spatial control. Edit by track produces intuitive, compelling edits using 3D point tracks.
https://edit-by-track.github.io/
GenWildSplat
Feedforward 3D models are awesome, but they can't handle in-the-wild images with varying illumination.
Check out GenWildSplat that fills this gap! https://genwildsplat.github.io/
TraceGen
Video world models help robot learning.
But, why do we need to predict all these pixels?
Let's model the world using 3D traces! 馃憠 computationally efficient 馃憠 embodiment-agnostic 馃憠 task-relevant
https://tracegen.github.io/
Edit by Track
Video editing requires precise spatial control. Edit by track produces intuitive, compelling edits using 3D point tracks.
https://edit-by-track.github.io/
Coupled Diffusion
So many awesome 2D diffusion models for a wide range of tasks. BUT, how can we get the best of both 2D and multi-view diffusion models?
Coupled diffusion shows a simple training-free approach to do so!
https://coupled-diffusion.github.io/
UniVerse
Most existing customization models either require costly per-concept optimization or clean reference images.
UniVerse enables decomposing and composing multiple visual concepts from unsegmented images
https://universe-personalization.github.io/
SIMPACT
VLMs have common-sense and semantic reasoning capabilities.
But, they suck at predicting physical consequences.
We equip VLMs with physical reasoning through simulation-in-the-loop world modeling!
https://simpact-bot.github.io/
UniVerse
Most existing customization models either require costly per-concept optimization or clean reference images.
UniVerse enables decomposing and composing multiple visual concepts from unsegmented images
https://universe-personalization.github.io/
GenWildSplat
Feedforward 3D models are awesome, but they can't handle in-the-wild images with varying illumination.
Check out GenWildSplat that fills this gap! https://genwildsplat.github.io/
CVPR 2026 was a blast! 馃コ
It was great meeting old and new friends and presenting our work.
summarized below if you missed it! 馃У