Sony Patents a Way to Stop 3D Object Decoding From Bogging Down
Every time a 3D scene loads on your headset or phone, the device has to reconstruct complex objects from encoded data, and that process can stall. Sony's new patent targets exactly that bottleneck.
What Sony's submesh buffer system actually does
Every time a 3D scene loads on a device, whether in a game, a virtual meeting, or an augmented-reality app, the hardware has to rebuild each object's shape from compressed data. That rebuilding step can slow everything down, especially when objects are large or complicated.
Sony's patent describes a system that organizes that work more efficiently. Instead of hunting through a stream of data to find the pieces of a 3D shape, the device pre-stores a lookup table in a dedicated memory slot called a buffer. A scene description (essentially a master list of everything in the scene) points directly to that buffer, so the device always knows exactly where to look.
The result is that each piece of a 3D shape, called a submesh, can be found and decoded quickly without wasted searching. Sony says this approach keeps decoding running at full speed rather than hitting slowdowns.
… a scene description generation unit that generates a scene description representing a scene configured by a 3D object, and sets, in the scene description, first reference information to a first buffer that stores submesh association information …
Translation: The system creates a scene layout that points directly to stored data parts for faster 3D rendering.
How the scene description links to submesh data buffers
The patent centers on a more organized handshake between a scene description and the raw data that makes up a 3D object.
A scene description is a structured file that defines what objects appear in a 3D scene and where. Sony's system adds a specific pointer inside that file, called first reference information, that points to a dedicated memory slot (a first buffer). That buffer holds submesh association information, which is essentially an index telling the decoder exactly which chunk of encoded geometry data corresponds to which piece of the overall mesh.
The decoding process then works like this:
- The scene description is read and the pointer to the buffer is found.
- The device reads the submesh association information from that buffer.
- Using that index, the device locates the specific data for each submesh (a portion of the full 3D shape).
- The full mesh is reconstructed from those submesh pieces.
The key efficiency gain is that the decoder never has to search blindly through a data stream. The address of every piece is known before decoding starts, so processing stays consistent and avoids the stalls that happen when a decoder has to pause and scan for what comes next.
… submesh association information for specifying data of a submesh that is a part of a mesh representing the 3D object is set in the scene description …
Translation: Specific pointers are added to the scene plan to easily locate individual pieces of a 3D model.
What this means for streaming 3D content and XR scenes
For users, this is about whether a 3D scene loads quickly and plays back without hitches. Slow mesh decoding shows up as pop-in (objects appearing late), stuttering, or dropped frames, and those problems get worse as scenes become more detailed. Devices with limited processing power, like standalone XR headsets or phones, feel these slowdowns the most.
Sony's run of 3D graphics and XR filings suggests the company is building infrastructure for richer spatial content, likely tied to its PlayStation VR2 ecosystem and future mixed-reality hardware. A more efficient decoding pipeline would be a foundational piece of that, even if it sits far below what users ever directly see or touch.
Sony's 20th filing we've tracked since July in the AR glasses race watchlist builds on earlier applications like keeping faces tracked off-camera and pointing cameras with your gaze.
The problem this patent attacks is real. 3D mesh decoding is a genuine bottleneck in spatial computing, and it scales badly: the more detailed the object, the harder it is to keep decoding smooth on consumer hardware. That cost shows up directly in user experience as stuttering or slow-loading scenes.
The approach here is modest but sensible. Reorganizing how a decoder finds its data, by pre-indexing submesh locations in a dedicated buffer, is an engineering housekeeping move rather than a fundamental rethink. It addresses a real inefficiency without requiring new hardware.
Whether this specific indexing method is distinct enough from existing 3D format standards (like glTF, which already uses buffer-pointer systems) to hold up as a patent claim is a separate question. As a practical engineering decision inside a codec pipeline, it looks like a reasonable, targeted fix for a specific class of slowdown.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
69 drawing sheets from US 2026/0260424 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →