Researchers at the Beijing Institute of Technology and Tsinghua University Institute for AI have developed a new form of 3D Gaussian Splatting with CGI & VFX modeling potential.
As covered in our Gaussian Splatting feature earlier this year, the emerging technology turns anything from objects to scenes into 3D models, without requiring mesh editing. Its hassle-free nature has already seen the process integrated into apps like Polycam and KIRI Engine.
Aiming to accelerate and streamline Gaussian Splatting, engineers in China have now come up with ActiveSplat, an autonomous system capable of 3D mapping whole areas in real time.
Without delving into technical detail, they’ve achieved this thanks to hybrid map representation – a combination of dense and sparse data capture that allows the system to ‘navigate’ areas and scan with high accuracy. In a recent paper, the team revealed that it’s not just capable of avoiding repetitive trajectories, but adjusting to capture tricky surfaces on the move.
According to the researchers, all this unlocks the ability to “explore unknown indoor environments with high fidelity” and “opens the door to robotic autonomy research.” Given Gaussian Splatting’s inherent texture capture capabilities, their technology could also expedite CGI and visual effects workflows, by rapidly providing artists with detailed sketches.
That’s not to say ActiveSplat is a future replacement for 3D modeling – far from it. Creating realistic visuals is not a simple process, it often requires numerous technologies.
But the industry is starting to consider alternatives to sketching from scratch. Professional 3D scanning, for example, is increasingly being used to enhance results. With Gaussian Splatting also capturing 3D artists’ imagination, ActiveSplat could be the next step in this digital capture journey that empowers their creativity, while improving studios’ bottom lines.
These are exciting times for 3D data capture in CGI! You can find out more about the team’s innovative approach in their research paper here.
