To be fair, this is an academic work and to bring it up to the standards of a polished product will require engineering.
For example, it is much slower than real time. I am sure that we can make it realtime but it will be a big lift for us. Commercial solutions also have lots of other software built around them that address the needs of specific industries.
What this work shows is that markerless and high accuracy can go together and that it can all be trained using ONLY synethetic data.
In my lab at MPI, we have now gone completely markerless. MAMMA is trained to work well when multiple people are interacting closely and captures hand motion — these are challenging for marker-based systems.
Imagine AI that can handle any adverse condition-rain,fog,trasparent and reflective surfaces, and more 🌧️🪞🍷
Our #ECCV2024 work leverages diffusion models to tackle challenging scenarios, boosting monocular depth estimation! 🚀 @mattpoggi@pierluigi92
https://t.co/EaEm77v8Bu
MASt3R can still manage the impossible case, I think it will always amaze me! This was my room in Seattle, note that I had to concentrate to avoid image overlap ;)
📢📢📢 4DiM: Controlling Space and Time with Diffusion Models
https://t.co/UGvNiI7Dry
This is NOT a video model
This is NOT a radiance field
This is a model that learns from space-time data (what your phone captures), and can generate “slices” of space-time.