Microsoft Corp v. RealD Spark LLC

1. Case Identification

• Case #: IPR2023-01046
• Patent #: 10,740,985
• Filed: June 16, 2023
• Petitioner(s): Microsoft Corporation
• Patent Owner(s): RealD Spark, LLC
• Challenged Claims: 1-14 and 16-19

2. Patent Overview

• Title: Generating Reference Data for Eye-Gaze Correction
• Brief Description: The ’985 patent discloses methods for generating reference data to adjust a digital representation of a head region, such as for eye-gaze correction. The core technique involves training a first machine-learning algorithm (MLA) on image data to generate a first set of "editing instructions," and then using a second MLA, which takes the same training data and the output from the first MLA, to generate a second, refined set of editing instructions.

3. Grounds for Unpatentability

Ground 1: Anticipation over Ganin - Claims 1-3, 5, 10, 12-14, and 18 are anticipated by Ganin.

• Prior Art Relied Upon: Ganin ("Deepwarp: Photorealistic image resynthesis for gaze manipulation," a 2016 publication).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Ganin discloses every limitation of the challenged claims. Ganin describes a system for gaze redirection using two distinct neural network modules: a "coarse warping" module and a "fine warping" module, which correspond to the claimed first and second MLAs. The system is trained on pairs of images showing eye appearance before and after gaze redirection. The coarse module generates a "coarse warping field" (the claimed "first editing instructions") which, along with the original input, is fed into the fine warping module. The fine module then generates a "residual flow" that refines the coarse field to produce a final warping field (the claimed "second editing instructions"). This two-stage, coarse-to-fine process for generating editing instructions (warping fields) from training data was asserted to map directly onto the limitations of independent claim 1.
  • Key Aspects: Petitioner contended that Ganin's "coarse warping module" and "fine warping module" are the claimed first and second MLAs, and that the "warping fields" they generate are the claimed "editing instructions."

Ground 2: Obviousness over Ganin in view of Liu - Claims 16 and 19 are obvious over Ganin in view of Liu.

• Prior Art Relied Upon: Ganin (a 2016 publication) and Liu ("Optical Flow and Principal Component Analysis-Based Motion Detection in Outdoor Videos," a 2010 publication).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner asserted that Ganin teaches all limitations of claims 16 and 19 except for providing the editing instructions in a "compressed representation." The ’985 patent identifies principal component analysis (PCA) as a form of compressed representation. Liu was cited for its teaching of applying PCA to compress two-dimensional optical flow data. Petitioner argued that Ganin's warping fields are a type of optical flow, making Liu's teachings directly applicable.
  • Motivation to Combine: A Person of Ordinary Skill in the Art (POSITA) would combine Ganin with Liu to improve Ganin’s system. Applying PCA, as taught by Liu, to Ganin’s warping fields would compress the data, reduce noise, and achieve speed optimizations, which Ganin itself suggests as an area for improvement. Both references are in the analogous art of image processing.
  • Expectation of Success: A POSITA would have a high expectation of success, as applying PCA to motion-related image data like optical flows was a well-known, standard, and predictable technique for data compression at the time of the invention.

Ground 3: Obviousness over Chalom in view of Aslan - Claims 1-3, 5-6, 8-14, and 16-19 are obvious over the combination of Chalom and Aslan.

• Prior Art Relied Upon: Chalom (Patent 10,664,949) and Aslan (Application # 2017/0132528).

• Core Argument for this Ground:

  • Prior Art Mapping: Petitioner argued that Chalom teaches a system for eye-gaze correction using a single MLA (a random forest classifier, or RFC) trained on pairs of source and target eye-region images to generate motion vector fields (editing instructions). Aslan teaches a "teacher-student" training architecture, where a larger, more complex "teacher" model is used to train a smaller, more efficient "student" model. In Aslan, the student model is trained using the original training data and the output from the teacher model. Petitioner contended that applying Aslan's teacher-student framework to Chalom's system renders the two-MLA structure of the challenged claims obvious.
  • Motivation to Combine: A POSITA would have been motivated to modify Chalom's system using Aslan's teacher-student architecture to gain efficiency and/or accuracy. For example, Chalom's trained RFC could serve as a "teacher" model to train a smaller, more compact "student" RFC, thereby improving computational performance—a goal explicitly mentioned in Aslan. Chalom also suggests that its random forest models could be compressed, which would have motivated a POSITA to look to known model compression techniques like Aslan's.
  • Expectation of Success: A POSITA would have a reasonable expectation of success in combining the references because the teacher-student paradigm was a well-known architecture for improving machine learning models in the analogous field of image processing.

• Additional Grounds: Petitioner asserted additional obviousness challenges for claim 4 based on the Chalom and Aslan combination in further view of Grau (Application # 2018/0158246), and for claim 7 based on the same combination in further view of Pace (Patent 8,942,283). These grounds relied on similar modification theories to add teachings of specific editing instruction types (e.g., brightness adjustment fields from Grau) or alternative compression techniques (e.g., wavelet transformation from Pace).

4. Arguments Regarding Discretionary Denial

• Petitioner argued that discretionary denial would be inappropriate.
• Under §325(d), it was asserted that although the Ganin reference was before the Examiner, the Patent Owner obscured its relevance during prosecution, leading the Examiner to err. The other asserted references were not considered during prosecution.
• Under §314(a) (and the Fintiv factors), Petitioner argued denial is unwarranted because it filed a Sotera stipulation, agreeing not to pursue in parallel district court litigation any ground raised or that reasonably could have been raised in the IPR.

5. Relief Requested

• Petitioner requests institution of an inter partes review and cancellation of claims 1-14 and 16-19 of the ’985 patent as unpatentable.