Patent Drafting Analysis of Google LLC’s Neural Rendering View-Dependent Images Using 2D Images | US 11,922,562 B2

IP Drafting Analysis · US 11,922,562 B2

Patent Drafting Analysis of Google LLC's Neural Rendering View-Dependent Images Using 2D Images | US 11,922,562 B2

Q: What does US 11,922,562 B2 protect?

US 11,922,562 B2 protects method claims for rendering view-dependent images of 3D objects from 2D image inputs. The patent addresses the problem that existing neural rendering approaches either achieve high image quality (NeRF) or real-time rendering (traditional pipelines) but not both simultaneously. The solution uses a signed distance function-based sinusoidal representation network (SDF-SIREN) to jointly model 3D shape via a zero-level set and appearance via a spatially varying emission function, then converts the resulting neural model to a triangular mesh compatible with traditional real-time graphics pipelines.

Q: Who owns patent US 11,922,562 B2?

US 11,922,562 B2 is owned by Google LLC, headquartered in Mountain View, CA, US. The inventors are Gordon Wetzstein (Fremont, CA), Andrew Jones (Fremont, CA), Petr Kellnhofer (Fremont, CA), Lars Jebe (Fremont, CA), Ryan Spicer (Fremont, CA), and Kari Pulli (Fremont, CA).

Q: What are the independent claims of US 11,922,562 B2?

Claim 1 is a method claim covering modeling 3D shape via a zero-level set of a signed distance function using a shape renderer, modeling appearance by minimizing image reconstruction error using an appearance renderer, combining them into a neural model, converting to a triangular mesh, and rendering at least one image using the triangular mesh. Claim 22 is a method claim similar to Claim 1 but explicitly recites modeling appearance based on a spatially varying emission function rather than by minimizing an image reconstruction error. Claim 24 depends from Claim 22 but also functions as an independent limitation, specifying that the image data is two-dimensional (2D) image data.

Q: What technology does this patent cover?

This patent covers a neural rendering technology that creates photorealistic 3D views of objects from regular 2D photographs, usable in real time. Traditional 3D computer graphics require manually created 3D models, while newer neural methods like NeRF create high-quality images but are too slow for real-time use. Google's patented approach trains a special neural network (using sinusoidal activation functions called SIREN) that simultaneously learns an object's shape and how it looks from different angles, then automatically converts this neural representation into a standard 3D mesh that can be rendered in real time on regular graphics hardware — bridging the gap between neural image quality and traditional rendering speed.

Q: What IPC classifications does this patent use?

G06T 15/20 (2011.01) — 3D image rendering with lighting or shading. G06F 3/01 (2006.01) — Input arrangements or combined input and output arrangements for interaction between user and computer. G06N 3/048 (2023.01) — Neural networks characterized by learning methods or architectures, specifically activation functions. G06T 7/55 (2017.01) — Depth or shape recovery from multiple images. G06T 17/20 (2006.01) — Free-form surface description using mathematical representation.

A structural and strategic analysis of US 11,922,562 B2, examining claim architecture, drafting quality, critical gaps, and prosecution positioning for Google LLC's SDF-SIREN neural rendering system.

US 11,922,562 B2Filed: Dec 14, 2021Granted: Mar 5, 2024G06T 15/20G06F 3/01G06N 3/048G06T 7/55G06T 17/20

Draft a Patent Like This Explore Related Patents

Spec Words

5,800

Across 6 sections

Draft now ↗

Total Claims

3 independent · 21 dependent

Draft now ↗

Figure Sheets

Display hardware, neural pipeline, system architecture, method flowchart, camera array

Draft now ↗

Published by PatSnap Insights Team · June 4, 2026 · 12 min read Verified by PatSnap Eureka Data

Overview

Structural Overview

The detailed description dominates at approximately 47% of total words (~3,200 of ~6,685 estimated words), with the claims section contributing a substantial ~24%, reflecting the mathematical depth of the SDF-SIREN rendering methodology. The patent presents 24 total claims across 3 independent claims — all of the method type — supported by 5 drawing sheets spanning display hardware, the neural network pipeline, system architecture, a rendering flowchart, and a custom camera array. Figure coverage is adequate for the core pipeline but notably absent for apparatus/system claim type alternatives.

Section Word Distribution

↗ Click bars to explore

Figure Inventory — 5 Sheets

Figure	Description	Role
FIG. 1A	Illustrates a light field display 100 comprising display area 102 formed by angular pixels and eye trackers 104 mounted at the top of the display.Search in Eureka ↗	Key embodiment
FIG. 1B	Illustrates a head mounted display 150 comprising left display 152 and right display 154 for presenting stereo view-dependent images based on head position.Search in Eureka ↗	Key embodiment
FIG. 2	Schematic of neural network system 200 showing 3D object 202, shape (SDF) renderer 204, appearance renderer 206, resulting 3D neural model 208, and output view-dependent images 210.Search in Eureka ↗	System architecture
FIG. 3	Block diagram of computing system 300 comprising processor 302, input/output 304, memory 306 with shape renderer 308, appearance renderer 310, and optional real-time renderer 312.Search in Eureka ↗	System architecture
FIG. 4	Flowchart 400 illustrating method steps: providing 2D images (402), modeling shape via zero-level set (404), modeling appearance via spatially-varying emission function (406), combining shape and appearance (408), converting to triangular mesh (410), and rendering view-dependent images (412).Search in Eureka ↗	Flow diagram
FIG. 5	Frontal view of custom camera array 500 showing six Back-Bone cameras 502 at center and sixteen background cameras 504 arranged on vertical rigs for capturing multi-view 2D images of subjects.Search in Eureka ↗	Other

Analysis powered by PatSnap Eureka. Patent text and figures publicly available from USPTO. Draft a Similar Patent

Claims

Claim Architecture Analysis

The patent contains 3 independent claims — Claims 1, 22, and 24 — all of the method type, meaning no system/apparatus or computer-readable medium claim type is present among the independents. The dependent:independent ratio is 7.00:1, which is slightly above the software/AI norm of approximately 4–6:1, suggesting reasonable fallback depth. Notably, the tripartite independent claim structure (Claims 1, 22, and 24) covers method variants but conspicuously omits system and CRM claim types, leaving significant enforcement coverage gaps for apparatus-based infringement.

Core inventive concept: The claims address the problem of achieving simultaneous high-quality novel view synthesis and real-time rendering from sparse 2D images — a capability neither NeRF nor IDR achieves alone. The mechanism, as expressed in Claim 1, is modeling a 3D shape via "a zero-level set of a signed distance function using a shape renderer" and modeling appearance via "minimizing an image reconstruction error" using an appearance renderer, then converting the resulting neural model into a "triangular mesh" for real-time rendering — combining implicit neural surface representation with traditional graphics pipeline compatibility.

Independent Claim Dissection

Claim	Preamble	Transition	Key Body Elements
Claim 1	A method	comprising	modeling 3D shape using image data via zero-level set of signed distance function with shape renderer, modeling appearance by minimizing image reconstruction error via appearance renderer, combining 3D shape and appearance to generate neural model, converting neural model to triangular mesh via rendering engine, rendering at least one image using triangular mesh and rendering engineSearch prior art ↗
Claim 22	A method	comprising	modeling 3D shape using image data via zero-level set of signed distance function with shape renderer, modeling appearance on spatially varying emission function via appearance renderer, combining 3D shape and appearance to generate neural model, converting neural model to triangular mesh, rendering at least one image using triangular mesh and rendering engineSearch prior art ↗
Claim 24	The method of claim 22	wherein	the image data is two-dimensional (2D) image data — narrows Claim 22 to specify input dimensionalitySearch prior art ↗

Claim Dependency Tree

1 Method: modeling 3D shape via SDF zero-level set, modeling appearance via image reconstruction error, converting to triangular mesh, rendering imagesSearch Claim 1 prior art ↗

2 Adds: signed distance function based at least in part on location in 3D space and first learnable parameter of sinusoidal representation networkSearch in Eureka ↗

3 Adds: obtaining zero-level set includes sphere tracing the signed distance functionSearch in Eureka ↗

4 Adds: sphere tracing includes defining view, projection matrix, ray origin, ray direction with vector normalization, iterative minimization, and solving for zero-setSearch in Eureka ↗

5 Adds: modeling appearance includes using a spatially varying emission functionSearch in Eureka ↗

6 Adds: defining spatially varying emission function for directions in a global coordinate systemSearch in Eureka ↗

7 Adds: conditioning spatially varying emission function by local normal direction via automatic differentiationSearch in Eureka ↗

8 Adds: spatially varying emission function includes second learnable parameter of sinusoidal representation networkSearch in Eureka ↗

9 Adds: optimizing parameters based on weights for at least one loss functionSearch in Eureka ↗

10 Adds: minimizing image reconstruction error for 3D object in foreground pixels of displaySearch in Eureka ↗

11 Adds: image reconstruction error based on RGB value of foreground pixel and pixels with RGB values and object masksSearch in Eureka ↗

12 Adds: regularizing signed distance function by eikonal constraintSearch in Eureka ↗

13 Adds: eikonal constraint based on portion of pixels with RGB values and object masksSearch in Eureka ↗

14 Adds: enforcing projected pattern to fall within at least one boundary of object masksSearch in Eureka ↗

15 Adds: enforcing projected pattern using soft mask loss for pixels other than foreground pixelsSearch in Eureka ↗

16 Adds: soft mask loss includes binary cross entropy and minimum value along entire ray approximated by dense samplingSearch in Eureka ↗

17 Adds: regularizing spatially varying emission function to avoid overfitting to training viewsSearch in Eureka ↗

18 Adds: regulating spatially varying emission function comprises linearizing angular behavior using a smoothness termSearch in Eureka ↗

19 Adds: rasterizing triangular mesh, projecting vertex positions to pixels, computing angles between rendering camera ray and projective texture map viewpointsSearch in Eureka ↗

20 Adds: applying unstructured lumigraph rendering to blend contributions from first textures sorted in ascending order using computed weightsSearch in Eureka ↗

21 Adds: at least one image is a view-dependent imageSearch in Eureka ↗

22 Method: modeling 3D shape via SDF zero-level set, modeling appearance via spatially varying emission function, combining, converting to triangular mesh, renderingSearch Claim 22 prior art ↗

23 Adds: modeling appearance performed after modeling 3D shape within a processing pipelineSearch in Eureka ↗

24 The method of claim 22, wherein image data is two-dimensional (2D) image dataSearch Claim 24 prior art ↗

Metric	This Application	Software / AI Industry Norm
Total claims	24	15 – 25
Independent claim count	3	2 – 4
Dependent : Independent ratio	7.00 : 1	4 – 8 : 1
Method claims present?	Yes — Claims 1, 22, 24	Common
System / apparatus claims?	No	Common

Analysis powered by PatSnap Eureka. Patent text and figures publicly available from USPTO. Draft a Similar Patent

Drafting Quality

Drafting Quality Signals

The claims exhibit strong technical specificity in their loss function and sphere-tracing limitations — Claims 4, 11, and 16 in particular add concrete mathematical fallback positions rarely seen in AI/ML patents — but the entire claim set is limited to method claims only, exposing Google to system and CRM design-arounds. The dependent claim chain from Claims 10 through 18 provides a well-articulated fallback ladder through the image reconstruction error pipeline, while the complete absence of apparatus and computer-readable medium claims represents a significant prosecution strategy weakness.

✅

Antecedent Basis

The claim language maintains clean antecedent basis throughout. For example, Claim 1 introduces "a shape renderer of a rendering engine" and subsequently references "the rendering engine" consistently. Similarly, Claim 5 introduces "a spatially varying emission function" and Claims 6–9 all reference "the spatially varying emission function" with proper antecedent. Claim 4's sub-elements (view, projection matrix, ray origin, ray direction) are introduced without article issues and are not carried into other claims unexpectedly, avoiding §112 antecedent problems.

Explore in Eureka ↗

✅

Spec–Claim Consistency

FIG. 4 (steps 402–412) directly maps to the ordered method steps of Claims 1 and 22, providing strong written description support. FIG. 2 supports the shape renderer (204) and appearance renderer (206) limitations of Claim 1. The mathematical formulations in the detailed description (equations 1–13) provide explicit support for Claims 11 (L_R loss), 12 (eikonal constraint L_E), 15–16 (soft mask loss L_M), and 17–18 (smoothness term L_S). FIG. 3 supports the system architecture referenced in several dependent claims but is not claimed as an apparatus, leaving a consistency gap between the richly described system and the method-only claims.

Explore in Eureka ↗

✅

Transition Word Usage

All independent claims (1, 22, 24) correctly use "comprising" as the transition, preserving open-ended claim scope and permitting infringement by systems that include additional elements beyond those claimed. The "comprising" choice is strategically sound for an AI/neural rendering method claim, as it prevents competitors from avoiding infringement by simply adding preprocessing or postprocessing steps. No narrowing transitions such as "consisting of" or "consisting essentially of" appear anywhere in the claim set, which is appropriate for this technology area.

Explore in Eureka ↗

✅

§112(f) Means-Plus-Function Risk

The claims avoid means-plus-function language throughout. Structural terms such as "shape renderer," "appearance renderer," and "rendering engine" are used rather than "means for rendering shape" or "means for modeling appearance," avoiding automatic §112(f) invocation. Claim 4's step-by-step sphere tracing recitation avoids "step for" language. The specification's FIG. 3 provides structural descriptions of the shape renderer 308, appearance renderer 310, and real-time renderer 312 as memory-implemented modules, further reinforcing structural interpretation of these terms in the event of any §112(f) challenge.

Explore in Eureka ↗

⚠️

§101 Eligibility Risk

Claims 1 and 22 face a moderate Alice Step 2A risk because the core concept — modeling 3D shapes and appearances via mathematical loss functions — could be characterized as an abstract mathematical concept under Alice Step 1. The §101 defense is provided by the hardware tie-in to "a shape renderer of a rendering engine" and "converting the neural model into a triangular mesh," grounding the claims in a specific technical implementation. However, without any system or CRM claims, if the method claims are rejected under §101, the entire grant is at risk — no alternative claim type exists as a fallback. The detailed description's computer-specific implementation details in FIG. 3 support an Enfish-style "improvement to computer functionality" argument.

Explore in Eureka ↗

✅

Dependent Claim Fallback Quality

The dependent claims from 2 through 21 provide a well-structured fallback ladder with genuinely distinct technical limitations. Claim 4's sphere-tracing sub-step breakdown (view matrix, ray origin, ray direction, iterative minimization) adds a unique computational pathway not captured in Claim 3. Claims 11–16 progressively narrow the image reconstruction error through RGB pixel values, eikonal constraint, mask enforcement, soft mask loss with binary cross entropy, which are distinct and non-redundant positions. The weakest dependent is Claim 21 ("the at least one image is a view-dependent image"), which adds minimal scope narrowing over Claim 1's already implicit rendering purpose.

Explore in Eureka ↗

⚠️

Abstract Quality

The abstract describes the general architecture but omits the key novel contribution — the joint optimization of shape (SDF-SIREN) and appearance (spatially varying emission function) via a combined multi-loss function (L_R + L_E + L_M + L_S). An examiner reading only the abstract would identify a neural network receiving 2D images and producing a triangular mesh, but would not identify the periodic sinusoidal activation function mechanism that differentiates this approach from NeRF and IDR. The abstract specifically mentions "signed distance function based sinusoidal representation network" but fails to mention the mesh export for real-time rendering compatibility, which is the primary commercial advantage of the invention.

Explore in Eureka ↗

⚠️

Figure Support Quality

FIG. 4 directly supports the core method claim steps, and FIG. 2 supports the shape/appearance renderer architecture. However, the mathematical loss functions claimed in Claims 11–18 (L_R, L_E, L_M, L_S equations) have no corresponding figure support — they appear only as inline equations in the specification text. The unstructured lumigraph rendering of Claim 20 (equations 11–13) similarly lacks a dedicated figure. FIG. 3's system block diagram supports Claims 19 and 20 implicitly via the real-time renderer 312, but a dedicated figure illustrating the rasterization pipeline and lumigraph blending would have provided stronger claim-to-figure correspondence for these downstream rendering limitations.

Explore in Eureka ↗

Analysis powered by PatSnap Eureka. Patent text and figures publicly available from USPTO. Draft a Similar Patent

Scorecard

Strategic Intent Scorecard

Multi-dimensional assessment of this application's patent strategy quality, based on claim structure, specification depth, and prosecution positioning.

Claim Breadth

3.5

Prosecution Defensibility

3.8

Spec–Claim Consistency

4.2

Dependent Claim Coverage

Claim Type Diversity

Figure Support Quality

3.5

Key observation: Spec–Claim Consistency scores highest (4.2/5.0) because the mathematical equations in the detailed description (equations 1–13) map precisely to claim limitations in Claims 2, 4, 8, 11–16, and 18–20, supported by FIG. 4's step-by-step flowchart alignment. Claim Type Diversity scores lowest (2.0/5.0) because all 24 claims — including all 3 independent claims — are method claims, leaving the entire apparatus/system architecture described in FIG. 3 and the computing system of the Summary section unclaimed as patent-protected subject matter. Practitioners should note that a continuation filing covering system and CRM claims using the existing specification would significantly strengthen Google's enforcement position for this technology.

See how your own draft compares — Open Eureka IP Drafting →

Critical Gaps

3 Critical Gaps in This Claim Set

A senior-attorney lens on the three highest-priority structural weaknesses — what each exposes in prosecution and litigation, and what a stronger filing would have done differently.

GAP 01 · HIGHEST IMPACT

Complete Absence of System and CRM Independent Claims

The structural weakness is that all 24 claims, including all 3 independent claims, are method claims — despite the specification fully describing a computing system (FIG. 3, processor 302, memory 306, shape renderer 308, appearance renderer 310, real-time renderer 312) and a head mounted display/light field display (FIGs. 1A, 1B) that implement the invention. This creates a critical design-around opportunity: a competitor manufacturing or selling hardware systems or distributing software incorporating the SDF-SIREN pipeline could argue no method is being "performed" by the manufacturer in a single jurisdiction, avoiding direct infringement of the method claims. A stronger filing would have included at minimum one system claim ("A system comprising: a processor; and a memory storing instructions that when executed by the processor cause the system to...") and one computer-readable medium claim, directly corresponding to the FIG. 3 embodiment already fully supported by the specification.

GAP 02 · HIGH IMPACT

Missing Claim Coverage for Real-Time Rendering Output to Display

Claims 1 and 22 claim only up to "rendering at least one image using the triangular mesh and using the rendering engine," without claiming the downstream display step — tracking head or eye position and generating left/right stereo views for a head mounted display or light field display. The specification (col. 5–8) extensively describes how views are generated based on eye tracking data (FIG. 1A, trackers 104) and head position data (FIG. 1B) and displayed on the left/right displays (152, 154), and the Summary explicitly includes "track head position of a viewer" and "display the left view to a left eye" as part of the system claim scope. This creates a gap where a competitor could implement the neural model generation step of Claim 1 and then use a different display/view-selection mechanism without infringing. A stronger filing would have added an independent claim covering the end-to-end pipeline through tracked-view display on a head mounted or light field display, as fully supported by the specification.

GAP 03 · HIGH IMPACT

No Claim Coverage for Training Data Capture Pipeline

Unlock to read the full analysis.

🔒

3 Critical Gaps in This Claim Set

See the full attorney-level analysis of what this application leaves unprotected — and how to draft it more defensively for your own filings.

No system or CRM independent claims Display output pipeline unclaimed Camera array capture pipeline unclaimed

Unlock Full Analysis — Free

Frequently asked questions

US 11,922,562 B2 — key questions answered

What does US 11,922,562 B2 protect?+

Who owns patent US 11,922,562 B2?+

What are the independent claims of US 11,922,562 B2?+

What technology does this patent cover?+

What IPC classifications does this patent use?+

Still have questions? PatSnap Eureka can answer them from patent data instantly. Search in Eureka

Disclaimer: This analysis is generated by PatSnap Eureka AI based on publicly available patent data from the USPTO. It does not constitute legal advice and should not be relied upon as such. Patent data may be subject to change as prosecution progresses. Scores and assessments reflect automated analysis and may not capture all relevant legal or technical nuances. Always consult a qualified patent attorney for formal legal opinions on patentability, freedom to operate, or infringement.

PatSnap Eureka

Draft Your Next Patent with AI

Eureka Patent Drafting Agent generates complete patent application drafts, including claims, specifications, abstracts, and figure descriptions. Designed for patent professionals and R&D teams, it supports efficient preparation of structured, filing-ready patent documents from technical invention disclosures.

Draft a Patent

Patent Information

Patent No.US 11,922,562 B2

AssigneeGoogle LLC

FiledDecember 14, 2021

PublishedMarch 5, 2024

PriorityDecember 14, 2020

App. No.17/644,291

StatusGranted · Active

Explore Further in Eureka

Neural implicit surface rendering patents› Google LLC rendering patent portfolio› NeRF competitor neural view synthesis› Sphere tracing signed distance function prior art› Light field display eye tracking rendering›

Ask anything about this patent.
PatSnap Eureka searches patents and data to answer instantly.

Link copied to clipboard

AI AGENTS

AI APPLICATIONS

OTHERS

INDUSTRIES

USE CASES

EXPLORE

ENGAGE

SUPPORT & SERVICES

Your Agentic AI Partner
for Smarter Innovation

Great, Please verify your email.

Patent Drafting Analysis of Google LLC’s Neural Rendering View-Dependent Images Using 2D Images | US 11,922,562 B2

Patent Drafting Analysis of Google LLC's Neural Rendering View-Dependent Images Using 2D Images | US 11,922,562 B2