Positive users highlight strong booth interest in the 3D point tracking video editing demo as proof of its value, while negative users dismiss the work as vanity metrics that fail to scale.
VLMs are great for their semantic reasoning capabilities.
BUT, they don't have a grounded understanding of physical dynamics.
So, we equip VLMs with physical reasoning through simulation-in-the-loop world modeling.
Come talk with Haowen about SIMPACT (#611): https://simpact-bot.github.io/
Customizing multiple visual concepts remains challenging.
We introduce UniVerse, a segmentation-free framework that learns to disentangle, localize, and recombine individual concepts within diffusion transformers.
Come check out the cool results by Quynh (#53). https://universe-personalization.github.io/
Robot learning is bottlenecked by data: adapting to new tasks, robots, and environments often requires many demonstrations.
Meet TraceGen! A motion world model that learns in a compact 3D trace space, abstracting away complex appearance.
This enables learning from diverse videos.
Meet @JayLEE_0301 at #605!
I will be around all poster sessions!
Please come say HI!
@jbhuang0604 @YaoChihLee Same old vanity metrics, wake me up when it actually scales.
@jbhuang0604 @YaoChihLee 378 booth visits and the line still long lol
worth it for 3D point tracking