Artificial Intelligence Industry Association Inc v. Parallel Domain Inc

I. Executive Summary and Procedural Information

• Parties & Counsel:
  • Plaintiff: Artificial Intelligence Industry Association, Inc. (Florida)
  • Defendant: Parallel Domain, Inc. (Delaware)
  • Plaintiff's Counsel: FRY Law Corp
• Case Identification: 3:25-cv-07658, N.D. Cal., 12/04/2025
• Venue Allegations: Plaintiff alleges venue is proper in the Northern District of California because Defendant has committed acts of infringement in the district, including selling its accused software products to customers with major operations in the district, such as Google, Continental, Woven Planet, and Toyota Research Institute.
• Core Dispute: Plaintiff alleges that Defendant's synthetic data generation platform infringes three patents related to generating synthetic images for training machine learning models and embedding calibration metadata into video files.
• Technical Context: The technology involves creating high-fidelity, virtual worlds and sensor data to train and test artificial intelligence systems, a critical component for advancing development in fields like autonomous driving.
• Key Procedural History: The complaint states that prior to filing, Plaintiff sent Defendant a formal demand letter identifying the asserted patents and alleging infringement. Plaintiff alleges that Defendant rejected an offer to license the patents and continued its allegedly infringing activities.

Case Timeline

Date	Event
2015-04-29	Priority Date (U.S. Patent No. 9,930,315)
2015-04-29	Priority Date (U.S. Patent No. 10,075,693)
2018-03-27	Issue Date (U.S. Patent No. 9,930,315)
2018-09-11	Issue Date (U.S. Patent No. 10,075,693)
2018-10-19	Priority Date (U.S. Patent No. 11,257,272)
2022-02-22	Issue Date (U.S. Patent No. 11,257,272)
2023-06-XX	Accused "Data Lab" API Announced
2025-12-04	Complaint Filing Date

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 11,257,272 - Systems and Methods for Generating Labeled Image Data for Machine Learning Using a Multi-Stage Image Processing Pipeline

The Invention Explained

• Problem Addressed: The patent's background describes the scarcity and high cost of acquiring the vast, specialized, and richly annotated image datasets required to train effective computer vision (CV) and machine learning models '272 Patent, col. 2:19-39 Manually capturing such data is time-consuming, expensive, and often results in low-quality or limited datasets '272 Patent, col. 2:33-39
• The Patented Solution: The invention provides an automated system for generating synthetic image data '272 Patent, col. 2:39-42 This system assembles virtual 3D scenes by combining elements from databases of background images, 3D models, and textures '272 Patent, Fig. 8 It then uses configurable virtual cameras, which can mimic the properties of real-world cameras, to capture various perspectives of the scene '272 Patent, col. 4:1-14 The output is not just a simple image but can include additional data channels like depth maps, segmentation data, and optical flow data, creating a comprehensive dataset for training machine learning systems '272 Patent, abstract '272 Patent, col. 6:21-52
• Technical Importance: This automated approach provides a scalable and cost-effective method to overcome the data acquisition bottleneck in AI development, enabling the creation of diverse, precisely controlled, and richly annotated training data required for sophisticated CV tasks '272 Patent, col. 2:48-51

Key Claims at a Glance

• The complaint asserts at least Claim 1 and Claim 17 Compl. ¶41 Compl. ¶49 The infringement analysis focuses on method claim 1.
• Essential elements of Independent Claim 1 include:
  • Receiving databases of background images, 3D models, texture materials, and camera setting files.
  • Constructing a synthetic image scene with a graphics rendering engine, which involves selecting and arranging a background image, selecting and arranging a 3D model, and covering the 3D model with a texture material.
  • Placing a virtual camera in the scene, defined by a camera settings file.
  • Rendering projection coordinates from the image plane as pixel coordinates of a synthetic image.
  • Appending the synthetic image to a training dataset having a common image scene class.
  • Wherein the training dataset is used for training a machine learning system to perform a computer vision task.
• The complaint does not explicitly reserve the right to assert dependent claims for the '272 Patent.

U.S. Patent No. 10,075,693 - Embedding Calibration Metadata Into Stereoscopic Video Files

The Invention Explained

• Problem Addressed: When playing back stereoscopic 3D video, particularly video combined from multiple different cameras, the playback device requires specific camera and sensor parameters to render the frames correctly '693 Patent, col. 1:44-50 Without a standardized way to associate these parameters with the correct video segments, accurate playback is difficult '693 Patent, col. 1:62-67
• The Patented Solution: The invention proposes embedding camera, sensor, and processing parameters (calibration data, IMU data, GPS data, etc.) directly into the video file in real-time as it is being recorded '693 Patent, col. 2:5-9 This is accomplished by encoding the metadata into channels within the video file format, such as subtitle or closed captioning fields, ensuring the metadata is time-sequenced with the corresponding video frames '693 Patent, abstract '693 Patent, col. 9:10-17 A playback device can then parse this embedded data to properly calibrate and display the video '693 Patent, Fig. 11
• Technical Importance: This method ensures that complex stereoscopic video content remains self-contained and portable, allowing for accurate playback on various devices without external parameter files, a key requirement for reliable virtual reality experiences '693 Patent, col. 2:9-14

Key Claims at a Glance

• The complaint asserts at least Claim 1 Compl. ¶128 The infringement analysis centers on system claim 6.
• Essential elements of Independent Claim 6 include:
  • A computer store containing a stereoscopic video feed from a capture device and a plurality of contemporaneous metadata feeds.
  • A computer processor programmed to:
  • Obtain the stereoscopic video feed and metadata feeds.
  • Parse the metadata feeds.
  • Calibrate the video feed for display on a virtual-reality headset using the metadata.
  • Decode metadata stored in subtitles or closed captioning fields of the video file format, with the timing of the metadata associated with the timing of the subtitle/captioning fields.
• The complaint does not explicitly reserve the right to assert dependent claims for the '693 Patent.

U.S. Patent No. 9,930,315 - Stereoscopic 3D Camera for Virtual Reality Experience

Technology Synopsis

The patent addresses the need for stable stereoscopic 3D video for virtual reality (VR) applications Compl. ¶45 It discloses methods for recording 3D video and embedding real-time calibration and motion data (e.g., from a gyroscope) to enable playback stabilization, which is critical for an immersive VR experience Compl. ¶45 Compl. ¶56

Asserted Claims

At least Claim 1 Compl. ¶116

Accused Features

The complaint alleges that Defendant's simulation systems infringe by using stereoscopic sensors to capture depth-aware data, embedding sensor metadata in real-time, and applying motion filtering and stabilization in its 3D-to-2D mapping processes Compl. ¶59

III. The Accused Instrumentality

Product Identification

The accused instrumentalities are Defendant Parallel Domain, Inc.'s software products and services, collectively referred to as its "synthetic data generation platform" Compl. ¶11 This includes specific components such as "simulation APIs, SDKs, and web tools," "PD Replica," the "PD SDK," and the "Data Lab API" Compl. ¶¶4-5 Compl. ¶10

Functionality and Market Context

• The platform is designed to programmatically generate high-fidelity synthetic data for training and testing computer vision and perception systems Compl. ¶¶46-47 It allows users to create virtual scenes from 3D assets, render realistic images with configurable virtual sensors, and generate corresponding data layers like depth maps and semantic segmentation Compl. ¶¶48-50 The complaint includes a screenshot from a YouTube video titled "Why You Need Synthetic Data" to demonstrate that creating synthetic images for machine learning training is central to Defendant's business model Compl. p. 2 The platform is marketed to major entities in the autonomous systems space, including Google and Toyota Research Institute, as a tool to "close the 'sim-to-real gap'" by improving AI model performance Compl. ¶18 Compl. ¶48
• Specific functionalities alleged in the complaint include a database of over 3,000 3D models, GPU-accelerated physics-based rendering, extensive camera configuration options via its PD-SDK, and the generation of training datasets that have led to documented performance improvements in machine learning models Compl. ¶19 Compl. ¶62 Compl. ¶65 Compl. ¶69

IV. Analysis of Infringement Allegations

U.S. Patent No. 11,257,272 Infringement Allegations

Claim Element (from Independent Claim 1)	Alleged Infringing Functionality	Complaint Citation	Patent Citation
A method for generating a synthetic image data by capturing a camera view of a synthetic image scene, the method comprising:	Parallel Domain's synthetic data generation platform generates synthetic images by capturing virtual camera views of procedurally generated 3D scenes.	¶48	col. 47:56-60
receiving, by a processor, a database of background images, an object database including 3D models, a library of texture materials, and a library of camera setting files;	The platform is alleged to maintain and use databases of background environments, a 3D model database with over 3,000 assets, a library of texture materials, and camera setting files via its SDK.	¶¶62-63	col. 27:25-34
constructing a synthetic image scene rendered by a graphics rendering engine;	The platform uses GPU-accelerated graphics rendering to create photorealistic synthetic scenes with advanced lighting and physics.	¶19; ¶62	col. 15:20-25
selecting a background image from the database and arranging the background image in a background portion of the synthetic image scene;	The PD Replica tool is alleged to generate digital twins from real-world captures, creating background environments with semantic segmentation, ground meshes, and depth maps.	¶63	col. 17:3-11
selecting a 3D model from the object database and arranging the 3D model in a foreground portion of the synthetic image scene;	The platform is alleged to select and place 3D models from its asset database, including vehicles and pedestrians, into foreground positions within scenes.	¶62; ¶123	col. 17:41-47
covering the 3D model with a texture material selected from the library of texture materials;	The platform is alleged to apply materials from a texture library to 3D models, with "technically infinite permutations through runtime color assignment and accessory variations."	¶62	col. 17:47-54
placing a virtual camera in the foreground portion of the synthetic image scene;	Virtual cameras are allegedly positioned in scenes using the SensorExtrinsic class, which controls position and orientation.	¶66	col. 18:29-35
rendering projection coordinates included in the image plane as pixel coordinates of a synthetic image;	The platform allegedly renders 3D to 2D projections using camera intrinsics, outputting RGB images and other data channels projected to a 2D image plane.	¶67	col. 18:55-65
appending the synthetic image to a training dataset comprising synthetic images having a common image scene class, wherein the training dataset is used for training a machine learning system to perform a computer vision task.	Generated images are allegedly organized into training datasets by scene class and used to train machine learning models, with documented performance improvements cited as evidence of this use.	¶¶68-69	col. 48:50-59

U.S. Patent No. 10,075,693 Infringement Allegations

The complaint does not provide a claim chart exhibit for the '693 Patent. The chart provided in Exhibit D for U.S. Patent No. 10,979,693 appears to be for a different, unasserted patent. The following analysis summarizes the narrative infringement theory for the '693 Patent presented in the body of the complaint.

The complaint alleges that Parallel Domain's platform directly infringes the '693 Patent by embedding calibration and sensor metadata into its generated synthetic video feeds Compl. ¶¶51-52 The theory centers on the allegation that the platform generates frames from multiple virtual cameras, computes the associated calibration data (both intrinsic and extrinsic), and stores this data alongside the image data in a "calibration" folder Compl. ¶54 Compl. ¶80 This process is described as embedding the metadata in "real-time" during generation Compl. ¶53 Compl. ¶79 Users are said to access this calibration data via the PD-SDK for downstream analysis, which the complaint contends constitutes a direct violation of the patent's claims related to parsing and using embedded metadata for calibration Compl. ¶54

• Identified Points of Contention:
  • '272 Patent - Use for Training: A potential point of contention for the '272 Patent may be the final step of the claim: "wherein the training dataset is used for training a machine learning system." The complaint provides evidence of performance improvements that result from using the synthetic data Compl. ¶¶69-72, which suggests training occurred. However, the analysis may focus on what level of proof is required to demonstrate that the data was actually used for training, as opposed to merely being created and provided for that purpose.
  • '693 Patent - "Embedding" Scope: A significant point of contention for the '693 Patent will likely be the definition of "embed." Claim 1 of the patent, as described in the complaint, requires embedding metadata "into the stereoscopic video feed...utilizing subtitle or closed captioning fields" Compl. ¶51 The complaint alleges that Parallel Domain stores calibration data in a separate "calibration" folder within a dataset structure Compl. ¶54 Compl. ¶80 This raises the question of whether storing metadata in an associated file within a directory structure can be considered equivalent to embedding it directly into the video stream's subtitle or captioning fields as the claim language specifies.

V. Key Claim Terms for Construction

• The Term: "embed... into the stereoscopic video feed... utilizing subtitle or closed captioning fields" (from '693 Patent, Claim 1, as recited in Compl. ¶51)
• Context and Importance: This term is the core of the technical mechanism in the '693 Patent. The infringement case hinges on whether Defendant's alleged practice of storing calibration data in an associated folder Compl. ¶54 Compl. ¶80 meets this limitation. Practitioners may focus on this term because there appears to be a mismatch between the specific technical implementation claimed and the one alleged to be infringing.
• Intrinsic Evidence for Interpretation:
  • Evidence for a Broader Interpretation: A party could argue that the patent's broader purpose is to ensure that metadata is inseparable from and time-synchronized with the video data to enable accurate playback '693 Patent, abstract They may suggest that "utilizing subtitle or closed captioning fields" is an exemplary, not exclusive, method of achieving this functional goal.
  • Evidence for a Narrower Interpretation: The claim language is highly specific, calling out "subtitle or closed captioning fields" as the mechanism for embedding data "into the stereoscopic video feed" '693 Patent, col. 9:15-17 The specification consistently describes embedding parameters "directly into the video file" '693 Patent, col. 2:5-6, which could support a narrow construction that excludes storage in separate, external files, even if they are in the same directory.
• The Term: "wherein the training dataset is used for training a machine learning system" (from '272 Patent, Claim 1)
• Context and Importance: This limitation requires an action to be performed with the created dataset, making it a crucial element for proving direct infringement of the method claim. The dispute may turn on whether providing data for the purpose of training and showing its effectiveness Compl. ¶¶69-74 satisfies the requirement that it is actually used for training.
• Intrinsic Evidence for Interpretation:
  • Evidence for a Broader Interpretation: The patent's abstract and background emphasize that the entire purpose of the invention is to create data "useful as training data for machine learning systems" '272 Patent, col. 2:48-51 An argument could be made that generating and appending the data to a "training dataset" with a specified "image scene class" inherently implies its intended and actual use, fulfilling the spirit of the claim.
  • Evidence for a Narrower Interpretation: The plain language of the claim requires the action of "using" the dataset for training. The specification describes the full pipeline, including the step where "machine learning systems may train one or more models" using the data '272 Patent, col. 7:62-65 A party could argue this is a distinct, required step that must be proven to have occurred, and that merely creating a dataset labeled for training is insufficient.

VI. Other Allegations

• Indirect Infringement: The complaint alleges both induced and contributory infringement. Inducement is alleged based on Defendant's provision of "detailed technical documentation, tutorials, and customer support services" that allegedly instruct customers to use the products in an infringing way Compl. ¶10 Contributory infringement is alleged on the basis that Defendant's products are material components "especially made for use in infringement" and lack substantial non-infringing uses Compl. ¶11
• Willful Infringement: Willfulness is alleged based on Defendant having received a "formal demand letter" prior to the lawsuit, which provided notice of the asserted patents Compl. ¶14 The complaint alleges that despite this notice and an offer to license, Defendant "elected to continue willfully violating" the patents Compl. ¶15 Compl. ¶111

VII. Analyst's Conclusion: Key Questions for the Case

• A core issue will be one of definitional scope: The case for the '693 patent may depend on whether storing calibration data in a separate folder within a dataset directory can be construed as infringing a claim that explicitly recites embedding metadata "into the stereoscopic video feed... utilizing subtitle or closed captioning fields." This presents a fundamental question of whether a functional similarity can overcome a specific structural limitation in the claim.
• A second key issue will be evidentiary: For the '272 patent, the infringement analysis will likely focus on what evidence Plaintiff presents to prove that the synthetic datasets generated by Defendant's platform were actually "used for training a machine learning system." The outcome may depend on whether evidence of improved model performance and customer testimonials is sufficient to meet this active-use requirement of the claim.
• A final question will concern the willfulness allegation: The court will examine the facts surrounding the pre-suit demand letter and licensing negotiations mentioned in the complaint. The determination of whether Defendant's continued alleged infringement was willful, despite notice, will be critical for the assessment of any potential enhanced damages.