PTAB

IPR2023-00924

Meta Platforms Inc v. VL IP Holdings LLC

Log In

1. Case Identification

2. Patent Overview

  • Title: Detection and Tracking of Objects in Images
  • Brief Description: The ’980 patent discloses a computer-implemented method for detecting and tracking objects in a series of images, such as a video. The method uses a "spatio-temporal model" that represents an object as a collection of components and determines the probability that the object is present in an image.

3. Grounds for Unpatentability

Ground 1: Obviousness over Zhao - Claims 1-16 are obvious over Zhao.

  • Prior Art Relied Upon: Zhao (Liang Zhao, "Dressed Human Modeling, Detection, and Parts Localization," a 2001 doctoral dissertation from Carnegie Mellon University).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Zhao, which was not considered during prosecution, discloses all limitations of the challenged claims. Zhao described an integrated system for human shape modeling, detection, and body part localization in images. Petitioner asserted that Zhao’s "human model," which encodes spatial relationships between body parts and incorporates temporal motion information from previous frames to resolve ambiguity, constitutes the claimed "spatio-temporal model." Further, Zhao's use of a "Bayesian Similarity Measure" to evaluate the resemblance between an image contour and the model meets the limitation of determining a probability that the object is in the image, which is then compared to a threshold for detection.
    • Key Aspects: This ground asserts that a single reference, which predates the patent's priority date by several years, renders the entire invention obvious without needing combination with other art.

Ground 2: Obviousness over Steffens and Zhao - Claims 1-16 are obvious over Steffens in view of Zhao.

  • Prior Art Relied Upon: Steffens (Patent 6,301,370) and Zhao.
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner contended that Steffens teaches a vision-based object detection and tracking system using a spatio-temporal model. Steffens’s model is a graphical model with nodes representing points on an object and edges representing spatial vectors between them; it also includes tracking these nodes between image frames, satisfying the temporal aspect. Steffens identifies objects based on a "similarity" measure between an image and a model but does not detail a specific probabilistic framework. Zhao was argued to supply this missing element with its explicit disclosure of a "Bayesian Similarity Measure," a robust probabilistic method for quantifying such similarity.
    • Motivation to Combine: A POSITA would combine Steffens with Zhao because both address the same problem of object detection using similar spatio-temporal graphical models. A POSITA implementing Steffens’s system, which relies on an undefined "similarity" measure, would have been motivated to look to well-known probabilistic frameworks like Zhao’s to implement this feature, thereby improving the system's precision and reliability.
    • Expectation of Success: A POSITA would have had a reasonable expectation of success in combining the references, as Zhao's probabilistic framework is a specific implementation of the general concept taught by Steffens. The combination represents applying a known technique (Zhao) to improve a similar, existing system (Steffens).

Ground 3: Obviousness over TLP - Claims 1-16 are obvious over TLP.

  • Prior Art Relied Upon: TLP (Leonid Sigal et al., "Tracking Loose-limbed People," a 2004 conference paper).

  • Core Argument for this Ground:

    • Prior Art Mapping: Petitioner argued that TLP, which was cited on an Information Disclosure Statement but never applied by the examiner, renders all challenged claims obvious. TLP discloses a probabilistic method for 3D human detection and tracking using a graphical model where nodes represent body parts and edges correspond to spatial and temporal relationships. This explicitly teaches the claimed "spatio-temporal model." TLP combines this body model with a "probabilistic image likelihood model" to infer the body pose, which Petitioner asserted is equivalent to determining a probability that the object is in each image. Detection is achieved by estimating belief in the graphical model, which would have been understood by a POSITA to involve comparing the determined probability against a threshold.
    • Key Aspects: Petitioner emphasized that this key reference was before the examiner but its teachings were never substantively considered during prosecution.

  • Additional Grounds: Petitioner asserted an additional obviousness challenge against claims 1-16 over Ozer (Patent 7,200,266) in view of Zhao. This ground relied on a similar motivation to combine as Ground 2, arguing a POSITA would have supplemented Ozer’s graph-matching algorithm with Zhao’s specific probabilistic similarity measure.

4. Key Claim Construction Positions

  • Petitioner presented no terms for construction but noted that the Patent Owner, in parallel litigation, appeared to construe "spatio-temporal model" as "a graphical model with nodes that correspond to each of the collection of components and to the object." Petitioner argued that all asserted grounds render the claims unpatentable even under this potential construction, contending that Zhao, Steffens, Ozer, and TLP each describe such a graphical model.

5. Arguments Regarding Discretionary Denial

  • Petitioner argued that discretionary denial under §325(d) is inappropriate because the petition raises new prior art and arguments not considered during prosecution. Specifically, Zhao and Steffens are new references. While TLP was cited, it was never applied or analyzed by the examiner, and the petition presents it with the benefit of expert testimony.
  • Petitioner also argued that denial under §314(a) based on Fintiv factors is unwarranted. The parallel district court case is in its infancy, with no answer filed, no discovery conducted, and no trial date set. Petitioner further stipulated that if the IPR is instituted, it will not pursue the same invalidity grounds in the district court action, mitigating concerns of duplicative efforts.

6. Relief Requested

  • Petitioner requests institution of an inter partes review and cancellation of claims 1-16 of the ’980 patent as unpatentable.