Sony Patents a Way to Compress 3D Scans Based on Their Own Detail Level
When you compress multiple 3D scans at once, treating them all the same wastes storage on low-detail data and degrades high-detail data. Sony's new patent wants to fix that by giving each scan its own compression setting.
How Sony's per-scan compression actually works
Today, most systems that compress collections of 3D scans apply the same compression settings to every scan in the batch, whether a scan is packed with fine detail or relatively sparse. That one-size-fits-all approach leaves file sizes larger than they need to be, or degrades the scans that matter most.
Sony's patent describes a system that looks at each 3D scan individually before compressing it. It checks how dense or how high-resolution that particular scan is, then picks a compression level that fits it. A richly detailed scan gets a finer-grained compression setting; a sparser scan gets a coarser one. The result is that each scan ends up as small as it can reasonably be without throwing away detail you'd actually notice.
For you, this matters whenever 3D scans are being stored, streamed, or processed in bulk, whether that's for AR headsets, self-driving car sensors, or medical imaging. Smaller files mean faster transfers and less storage cost, without a visible hit to quality.
an input receiving unit that receives an input of a plurality of pieces of point cloud data; a determination unit that determines a spatial quantization value of each of the plurality of pieces of point cloud data based on at least one of a resolution and a density of each of the plurality of pieces of point cloud data …
Translation: The system looks at the detail level of multiple 3D scans to figure out how much compression they need.
How the encoder picks a quantization value per scan
A point cloud is a collection of thousands or millions of 3D coordinates that together represent a physical object or environment, the kind of data a LiDAR sensor or depth camera produces. Compressing point clouds efficiently is genuinely hard because they are not uniform: some scans are dense and detailed, others are sparse.
Sony's patent describes an encoding apparatus with three main parts:
- Input receiving unit: accepts multiple point cloud datasets at once.
- Determination unit: analyzes each dataset's resolution (how fine-grained the grid of points is) and density (how many points fall in a given area), then assigns a spatial quantization value to it. Quantization, in this context, is essentially deciding how precisely to record each point's position: a high quantization value preserves more precision; a low one rounds positions more aggressively to save space.
- Encoding unit: compresses each dataset using the quantization value the determination unit picked for that specific scan, rather than a global default.
The key insight is that the quantization value is chosen per dataset, adaptive to that dataset's own characteristics. A high-density scan of a face, for example, would get a finer quantization value than a low-density scan of an open field. This avoids wasting bits on detail that was never there, and avoids crushing detail that was.
… an encoding unit that encodes each of the plurality of pieces of point cloud data based on the spatial quantization value determined by the determination unit.
Translation: It then compresses each 3D scan according to that calculated detail level.
What this means for 3D scanning and spatial computing
For anyone building or using systems that handle 3D data in bulk, such as AR and VR platforms, autonomous vehicles, or 3D mapping tools, encoding efficiency directly affects how fast data moves and how much it costs to store. A system that blindly applies the same compression to every scan either bloats file sizes unnecessarily or introduces artifacts in the scans that deserved better treatment.
Sony Semiconductor's steady investment in spatial-data processing fits a broader picture of the company positioning its sensor and chip divisions for LiDAR-heavy applications. Whether this specific approach makes it into a shipping product depends on a lot more than the patent itself, but the underlying problem it targets is real and affects every engineer working with point cloud pipelines today.
That makes this Sony's second filing we've tracked in Memory & storage since August, following a faster-recovery LiDAR sensor.
When a 3D scanning app or AR headset compresses the spatial data it captures, a one-size-fits-all approach degrades the result: some scenes come out bloated, others lose fine detail. Sony's patent tackles that by letting the compression system read each scan's resolution and density, then tune itself accordingly before storing or transmitting the data.
The person who benefits never sees the adjustment happening. They just get faster load times, smaller files, and sharper detail in the final 3D environment or scanned object.
That is a modest but real improvement. If you have ever watched a 3D model load slowly or noticed it look oddly smeared, uneven compression is a plausible culprit. Fixing it at this level means fewer of those moments reaching the user at all.
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
26 drawing sheets from US 2026/0268527 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →