Meta demonstrates SqueezeMe: Render 3 high-fidelity full-body avatars locally on Quest at 72 FPS

Source:Internet
2025-08-23 01:34:49

This past March, Meta demonstrated an avatar generation model called LRM (Large Reconstruction Model), which can create high-fidelity virtual avatars using smartphone selfie videos combined with cloud GPU rendering, with the underlying application of Gaussian splatting technology.


Recently, Meta researchers published a paper titled SqueezeMe: Mobile-Ready Distillation of Gaussian Full-Body Avatars, which introduces the team’s latest achievements in virtual avatars.


Researchers stated that by properly utilizing GPUs and NPUs, SqueezeMe can render 3 full-body avatars locally on the Quest 3 at 72 FPS. More importantly, there is almost no quality loss compared to the version rendered on a PC.


However, the data for these avatars needs to be captured by traditional capture arrays. These facilities are often equipped with more than 100 cameras and hundreds of lighting fixtures, and are quite expensive. Additionally, the device uses flat lighting and does not currently support lighting adjustments.


It remains unclear when SqueezeMe will be officially launched and applied.


Other News

  • Burnley Football Club will host its first VR live stream

  • China Telecom launched the Tianyi AI glasses, powered by the Qualcomm Snapdragon AR1 chip

  • Khronos Group and companies including Niantic Spatial are standardizing 3D Gaussian splatting in the glTF format