US 7436980

I'll search for authoritative information on this specific patent number before compiling the summary.

Let me check the IPR outcome and the second Federal Circuit appeal, which the user's "2026 docket" reference may implicate.

Let me verify the IPR outcome for this patent and check the February 2026 Federal Circuit matter that surfaced.

US Patent 7,436,980 (given as "7436980") — Concise Summary

Verification note: I searched for the exact number 7,436,980. Results consistently return this patent (title "Graphical Object Models for Detection and Tracking," App. No. 11/135,210), so I did not substitute any similar-looking number.

Abstract (verbatim)

"A computer implemented method for object detection includes providing a spatio-temporal model for an object to be detected, providing a video including a plurality of images including the object, and measuring the object as a collection of components in each image. The method further includes determining a probability that the object is in each image, and detecting the object in any image upon comparing the probabilities for each image to a threshold for detecting the object."

Plain-language overview of the independent claims

The patent has two independent claims, both covering the same underlying method — one drafted as a method, the other as a computer-readable medium.

Claim 1 — Computer-implemented method for object detection. The steps are:

1. Provide a spatio-temporal model of the object to be detected (i.e., a model that captures both the object's spatial structure and how it changes over time).
2. Provide a video made up of multiple images that include the object.
3. Measure the object as a collection of components in each image — rather than treating the object as one monolithic blob, it is decomposed into parts (e.g., for a pedestrian: head, left arm, right arm, legs; for a vehicle: its four corners).
4. Determine a probability that the object is present in each image (from the specification, this reasoning is done via message passing/belief propagation over the graphical model).
5. Detect the object in any given image by comparing the computed probabilities for each image against a detection threshold.

Claim 9 — Computer-readable medium embodying a computer program that performs the same five method steps above. (Note: claim 9's preamble recites "computer readable medium," though the subsequent dependent claims 10–16 confusingly recite "The method of claim 9.")

Dependent claims (brief)

• 2 / 10 — The spatio-temporal model includes a detector for each component.
• 3 / 11 — The model is a graphical model with nodes for each component and for the whole object.
• 4 / 12 — Probability for a current image is determined using component measurements taken from a prior image and a later image relative to the current one (forward-backward temporal reasoning).
• 5 / 13 — A temporal window of multiple images whose component measurements are passed forward to the current image.
• 6 / 14 — The probability covers the object's position and size.
• 7 / 15 — The detection threshold is determined empirically.
• 8 / 16 — Recites a specific joint probability distribution formula for the model with N components (factorized product of ψ potential terms between object nodes and component nodes, and φ local-evidence terms).

Enforcement history relevant to "docket" questions

PTAB — IPR2023-00924 (Meta Platforms, et al. v. VL IP Holdings LLC), filed May 24, 2023, FWD Dec. 5, 2024. Challenged claims 1–16 on §103 obviousness grounds (primary reference Zhao, "Dressed Human Modeling, Detection, and Parts Localization"). Per the Final Written Decision, the Board determined:

• Unpatentable: claims 1–3, 5–7, 9–11, 13–15
• Not shown unpatentable: claims 4, 8, 12, 16 (the claims reciting the prior+later-image measurement limitation and the specific joint-probability formula)

An inter partes review certificate (K1) issued December 19, 2025 (effective date 20251219), consistent with the FWD becoming final.

Federal Circuit appeals:

• 25-1415 (VL Collective IP, LLC v. Meta Platforms, Inc. et al.) — voluntarily dismissed per order dated August 11, 2025, deconsolidated from 25-1454.
• 25-1454 (Giphy, Inc. / Meta entities cross-appeal) — reported as dismissed (voluntary) in October 2025, with no merits ruling on validity.

On the "CAFC 2026 docket" request — limited/uncertain finding: The only 2026 Federal Circuit item I surfaced is a listing for "VL COLLECTIVE IP, LLC, Appellant v. UNIFIED PATENTS, LLC, Appellee" (Fed. Cir., Feb. 20, 2026). I cannot confirm that this 2026 appeal concerns U.S. 7,436,980 — it is not established by the sources retrieved. The confirmed Federal Circuit matters tied specifically to this patent are the two 2025 appeals above (25-1415 and 25-1454), both now dismissed. I also did not retrieve a live, docket-verified PACER record for any 2026 appeal of this patent; treat any 2026 date as unverified.

Uncertainties / caveats

• Claim survival after the IPR certificate: The FWD cancelled independent claims 1 and 9 while leaving dependent claims 4, 8, 12, and 16 (which depend from 1 and 9) not shown unpatentable. Because a dependent claim cannot stand alone when its parent is cancelled, the effective scope of the surviving claims after the K1 certificate is legally awkward and I do not have authoritative text of the certificate to state exactly which claims remain enforceable. Verify against the USPTO IPR certificate before relying on it.
• Legal status: Google Patents shows "Active, expires 2027-04-14," which predates the Dec. 2025 certificate; that status label should be treated as possibly stale.
• Assignee chain: Multiple recorded assignments (Siemens → Siemens Medical Solutions → Siemens Healthcare → VideoLabs → VL IP Holdings → VL Collective IP), plus a Praetor Fund I security interest later released. The current listed assignees are VL Collective IP LLC and VL IP Holdings LLC.

Sources: Google Patents US7436980B2 (authoritative full text supplied); USPTO assignment records (legacy-assignments.uspto.gov, reel 052142/frame 0403); Docket Alarm / Patently-O copy of the IPR2023-00924 Final Written Decision (patentlyo.com/media/2025/04/IPR2023-00924-FD-ready-for-mailing.pdf); CAFC order in 25-1415 (cafc.uscourts.gov, Aug. 11, 2025; law.justia.com/cases/federal/appellate-courts/cafc/25-1415/).

I'll research litigation involving US patent 7,436,980 using multiple sources.

I have strong leads. Let me dig into the Delaware case and the IPR, and check for any other lawsuits.

Litigation Involving U.S. Patent No. 7,436,980

Patents-in-suit identifier used: US 7,436,980 (also styled the "'980 patent"), titled "Graphical object models for detection and tracking," issued Oct. 14, 2008, original assignee Siemens Corporate Research, Inc. Current asserted owner of record in the litigation below: VL Collective IP, LLC (a VideoLabs affiliate; the patent passed Siemens → VideoLabs → VL IP Holdings LLC → VL Collective IP LLC).

All known litigation/enforcement proceedings involving the '980 patent are directed against the Meta family of companies. I found no other lawsuits asserting this specific patent.

1. District court action — the primary infringement suit

Details: VideoLabs asserted five patents in this case: U.S. 7,769,238; 8,139,878; 7,970,059; 7,266,682; and 7,436,980. Meta moved to dismiss the '682 and '980 patents under 35 U.S.C. § 101 (Alice). In a ruling dated April 22, 2024, the court granted the motion as to the '682 patent (abstract idea, no inventive concept) but denied it as to the '980 patent, holding the '980 claims patent-eligible under § 101. Sources:

• CourtListener opinion: https://www.courtlistener.com/opinion/[9636643](/patent/9636643)/videolabs-inc-v-meta-platforms-inc/
• IP Law Daily (Apr. 23, 2024): https://www.vitallaw.com/news/patent-d-del-mixed-results-for-meta-in-videolabs-ongoing-patent-infringement-lawsuit/ipm016afd26ab31c54f9cb17776126af43b5a
• Complaint (D. Del. 1:22-cv-00680): https://ai-lab.exparte.com/documents/dct/[4315675](/patent/4315675)/1-22-cv-00680_ded_complaint_exparte.pdf

Per Law360 (Oct. 2, 2025), the parties resolved the Delaware dispute ("Meta, VideoLabs Resolve Video Tech Patent Dispute"), and the Federal Circuit appeals were dismissed in Aug.–Oct. 2025.

2. PTAB — Inter Partes Review (IPR)

Sources:

• https://www.docketalarm.com/cases/PTAB/IPR2023-00924/Meta_Platforms_Inc._v._VL_IP_Holdings_LLC/
• https://ai-lab.exparte.com/case/ptab/IPR2023-00924/meta-platforms-inc-v-vl-ip-holdings-llc

(A companion IPR, IPR2023-00923, concerns U.S. 7,266,682 B2 — a different patent — so it is noted only for context.)

3. Federal Circuit appeals (from IPR2023-00924)

Federal Circuit order (Aug. 11, 2025) — deconsolidating and dismissing 2025-1415:

• https://www.cafc.uscourts.gov/08-11-2025-25-1415-vl-collective-ip-llc-v-meta-platforms-inc-order-25-1415-order-8-11-2025_2556477/
• https://law.justia.com/cases/federal/appellate-courts/cafc/25-1415/25-1415-2025-08-11.html

The dismissals were without a merits ruling (no affirmance/reversal of the IPR invalidity holding on the merits).

⚠️ Important caveat on the '980 patent's current validity status

The record here is internally inconsistent, and I am flagging it rather than resolving it:

• The PTAB Final Written Decision (IPR2023-00924) reportedly held the challenged claims unpatentable, and the U.S. Patent Office has issued an Inter Partes Review Certificate for the '980 patent (kind code K1, effective Dec. 19, 2025). An IPR certificate typically reflects cancellation or confirmation of claims in light of the final outcome — if the claims were held unpatentable and that holding became final, the certificate would ordinarily cancel them.
• Yet the Google Patents record still lists the patent as "Active, expires 2027-04-14," and a secondary source (PatSnap) describes the patent as "surviving" the appellate challenge because the appeals were dismissed without a merits ruling.

These cannot both be fully correct as to claim status. I recommend verifying directly via USPTO PatentCenter / PTAB (IPR2023-00924 final written decision and the K1 certificate text) rather than relying on any single aggregator, since AI-generated litigation summaries (e.g., the PatSnap page) may be inaccurate here. If you need this pinned down for an opinion or portfolio decision, that verification is essential.

Other VideoLabs enforcement activity (for context — '980 patent NOT asserted)

VideoLabs ran a broader 2022–2024 campaign, but the other suits I found did not assert the '980 patent, so they are outside your request (listed only to confirm I checked):

• VideoLabs, Inc. and VL Collective IP, LLC v. Netflix, Inc., No. 1:22-cv-00229 (D. Del.)
• VideoLabs v. Hisense, No. 2:24-cv-00904 (E.D. Tex.)
• VideoLabs v. Roku, No. 1:23-cv-01136 (D. Del.); VideoLabs v. ASUSTeK, No. 6:23-cv-00640 (W.D. Tex.)

None of these pleadings (as surfaced) identify U.S. 7,436,980 as an asserted patent.

Summary answer

The only litigation known to involve US 7,436,980 is:

1. VideoLabs, Inc. / VL Collective IP LLC v. Meta Platforms, Inc., et al., D. Del. 1:22-cv-00680-JHS, filed May 24, 2022 — Meta's § 101 motion denied as to the '980 patent (Apr. 22, 2024); case resolved/settled Oct. 2025.
2. IPR2023-00924 (Meta/Instagram/WhatsApp/Meta Technologies/Giphy v. VL Collective IP), filed May 24, 2023 — FWD Dec. 5, 2024 finding claims unpatentable; IPR certificate issued Dec. 19, 2025.
3. Federal Circuit appeals 2025-1415 and 2025-1454 — both voluntarily dismissed (Aug. 11, 2025 / Oct. 14, 2025) without a merits ruling.

If you are relying on the patent's current enforceability, verify the K1 IPR certificate and claim status directly at the USPTO, given the discrepancy noted above.

I'll reconcile the "no PTAB proceedings" structured block against the IPR evidence embedded in the patent text, then fill in claim-level details via search.

Let me verify the appeal dispositions, the certificate, and whether any second IPR exists on this patent.

⚠️ Contradiction flag (read first)

The structured "PTAB proceedings on file" block in this prompt says: "The USPTO ODP API returns no AIA trial proceedings for this patent." That is wrong for US 7,436,980. This patent was the subject of a fully-litigated IPR — IPR2023-00924 — that reached a Final Written Decision, spawned two Federal Circuit appeals, and produced an IPR certificate (K1) in December 2025. The evidence is in three independent places: (1) the authoritative patent text supplied in this prompt ("PTAB case IPR2023-00924 filed (Final Written Decision)"; "inter partes review certificate" legal event dated 2025-12-19); (2) the public FWD itself; and (3) the Federal Circuit orders below.

Per the operating rules, I treat the search/authoritative-text evidence as controlling and flag the ODP block as a stale/missed ingest — not as evidence of "no PTAB activity." Any advice built on the ODP "no proceedings" line would be dangerously wrong.

Separately, this resolves the uncertainty flagged in the prior summary: the February 20, 2026 Federal Circuit case VL Collective IP, LLC v. Unified Patents, LLC (No. 24-1890) is NOT about this patent. It arises from IPR2022-01086 on U.S. Patent No. 8,605,794 (VideoLabs patent). See below.

Proceedings overview

There is one (1) AIA trial proceeding on US 7,436,980: IPR2023-00924, status claims invalidated (FWD issued 2024-12-05, certificate 2025-12-19) — twelve of sixteen claims were held unpatentable, including both independent claims (1 and 9); four dependent claims (4, 8, 12, 16) were "not shown" unpatentable but depend solely from the now-canceled independent claims. Bottom-line defensive posture: the patent has been gutted, not "hardened" — every independent claim is canceled, so any demand letter citing claims 1, 2, 3, 5, 6, 7, 9, 10, 11, 13, 14, or 15 demands royalties on dead claims; the only formally surviving claims are inoperative orphans whose parents the PTAB canceled.

IPR2023-00924 — Meta Platforms, Inc., Instagram, Inc., WhatsApp LLC, Meta Platforms Technologies, LLC, and Giphy, Inc. v. VL Collective IP, LLC (PTAB caption also shows "VL IP Holdings LLC") — Patent 7,436,980 B2

• Type: Inter Partes Review
• Filed: 2023-05-24
• Status: Claims invalidated (Final Written Decision; terminated 2024-12-05; IPR certificate issued / effective 2025-12-19). Plain English: the Board canceled the independent claims and most dependents; a certificate has issued embodying that result.
• Judge panel: Administrative Patent Judges Karl D. Easthom, Jeffrey S. Smith, and David C. McKone. The decision includes a dissenting opinion — the dissent rejects the majority's inference that Zhao teaches comparing probabilities to a threshold ("We agree with Patent Owner for the reasons given by Patent Owner… The Majority adopts this theory"). (I could not confirm from the retrieved excerpt which APJ authored the dissent; the FWD PDF should be checked for the signature block before you rely on panel attribution for the dissent.)
• Petition grounds: All grounds were 35 U.S.C. § 103(a) against claims 1–16, on four theories per the FWD's grounds table: (1) Zhao (Liang Zhao, Dressed Human Modeling, Detection, and Parts Localization, CMU doctoral dissertation, 2001) alone; (2) Steffens + Zhao; (3) Ozer + Zhao; and (4) TLP (Leonid Sigal et al., Tracking Loose-limbed People, 2004). No § 102 or § 112 grounds. (Note: "TLP" is a Sigal-authored paper — i.e., the patentee's own prior work — and was on the IDS but never applied by the examiner.)
• Institution decision: Instituted 2023-12-06. The Board's institution analysis (cited throughout the FWD as "Inst. Dec. 23, 25") construed "spatio-temporal model" broadly in claim 1, holding it is not limited to a graphical model and "encompass[es] at least a model that encodes spatial relationships between parts of the object and incorporates temporal information used to detect the moving object, such as motion information." That construction — which read claim 3's "graphical model" limitation out of claim 1 — was the hinge of the whole case and the reason the § 103 grounds landed. (The FWD resolves all of claims 1–16, consistent with full institution.)
• Final Written Decision (2024-12-05; under 35 U.S.C. § 318(a) and 37 C.F.R. § 42.73): The Board determined, by a preponderance:
  • Unpatentable: claims 1, 2, 3, 5, 6, 7, 9, 10, 11, 13, 14, and 15 (verbatim from the FWD: "claims 1–3, 5–7, 9–11, and 13–15 are unpatentable"). This includes both independent claims — 1 (method) and 9 (computer-readable medium) — plus their like-numbered dependents.
  • Not shown unpatentable: claims 4, 8, 12, and 16. Reasoning, briefly:
    • Claims 4 and 12 — the "prior image and a later image relative to the current image" (forward-backward) limitation. The Board held Petitioner's expert "does not provide testimony that would support the Majority's inference," and that Zhao's bare statement that motion from previous frames "can be used to predict the orientations of the limbs" does not teach detecting a current-image object from prior and later-image measurements. "We find that Petitioner has not shown, by a preponderance of the evidence, that Zhao teaches 'detecting the object in a current image according to measurements of the object as a collection of components determined from a prior image and a later image relative to the current image' as recited in claim 4."
    • Claims 8 and 16 — the specific joint probability distribution formula. "We find that Petitioner has not shown, by a preponderance of the evidence, that Zhao would have rendered claim 8 obvious," agreeing with Patent Owner's expert that the prior-art formulas "would have to be rewritten in order to be like that of claim 8."
• Settlement / termination: No settlement. The proceeding ran to a contested FWD after an in-person oral hearing on 2024-09-10 (60 minutes per side, Hearing Room D, USPTO Alexandria). Petitioner was represented by Kirkland & Ellis LLP; Patent Owner by Reichman Jorgensen Lehman & Feldberg LLP.
• Appeal: Both sides appealed, then both walked away — no merits ruling.
  • 25-1415 (VL Collective IP, LLC v. Meta Platforms, Inc., et al.) — Patent Owner's appeal, from IPR2023-00924. Voluntarily dismissed 2025-08-11 (Fed. R. App. P. 42(b)(1)), deconsolidated from 25-1454. CAFC order · Justia
  • 25-1454 (Giphy, Inc. v. VL Collective IP, LLC) — original cross-appeal by the Meta/Giphy group (filed/docketed 2025-02-18); Meta, Instagram, WhatsApp, and Meta Platforms Technologies withdrew, leaving Giphy as the sole appellant; dismissed 2025-10-14 under FRAP 42(b), mandate issued same day. CAFC order · Justia
  • Net effect: the FWD is unreviewed — no Federal Circuit precedent, affirmance, or reversal. The Board's cancellation stands by operation of the certificate.
• Defensive value: High and lopsided. Any infringement theory resting on claim 1 or claim 9 is built on canceled claims — asserting them invites Rule 11/sanctions exposure, and a demand letter citing claims 2, 3, 5, 6, 7, 10, 11, 13, 14, or 15 cites claims the PTAB also canceled. The four live claims (4, 8, 12, 16) lost their independent parents, so they cannot be enforced as written without a reissue/re-examination to restructure the claims. Practically, this patent is a poor vehicle for assertion post-December 2025.

Strategic summary

Claim-by-claim status of US 7,436,980.

The "orphan dependent" problem is the whole ballgame. The Board's text ("Petitioner has not shown … that claims 4, 8, 12, and 16 are unpatentable") is a non-invalidation of the claim as written, not a ruling that a standalone invention is valid. Because claim 4 depends from claim 1, claim 8 from claim 1, claim 12 from claim 9, and claim 16 from claim 9, and because claims 1 and 9 were canceled, those four claims have no enforceable antecedent structure as the patent currently stands. A defendant should demand that VL identify the specific statutory vehicle by which it claims 4/8/12/16 are presently enforceable (e.g., a reissue under 35 U.S.C. § 251, or a certificate correcting the dependency) — absent that, they should be treated as unassertable. Verify this against the actual IPR certificate text before finalizing an opinion of counsel; I have the certificate's existence and date from the patent record, not its verbatim claim listing.

Estoppel landscape. Under 35 U.S.C. § 315(e)(2), the petitioners and their privies — Meta Platforms, Instagram, WhatsApp LLC, Meta Platforms Technologies, and Giphy — are estopped from raising in district court (and in ITC/§ 337 practice) any ground they raised or reasonably could have raised in IPR2023-00924. That locks out the four asserted combinations (Zhao; Steffens + Zhao; Ozer + Zhao; TLP) as invalidity defenses for that group. For a NEW defendant (non-petitioner, non-privy), no IPR estoppel attaches — the field is open on § 102/§ 103, including art not in the petition, and including the FWD's own reasoning (e.g., a later-image/forward-backward reference to attack the surviving claim 4/12 limitation, and a joint-probability-distribution reference to attack claim 8/16). One strategic caveat: because the Board rejected those grounds as to claims 4/8/12/16, VL can point to a favorable (though non-precedential, non-appealed) administrative record on those two limitations.

Pattern signals. This was a coordinated portfolio strike, not a one-off: the same petitioner group filed IPR2023-00923 against sibling U.S. Patent 7,266,682 the same day, before the same panel (Easthom, Smith, McKone), consolidated for a single 2024-09-10 hearing. So Meta/Giphy attacked the VideoLabs graphical-object-model patent and a coding patent in parallel, mirroring the D. Del. 1:22-cv-00680 action (VideoLabs, Inc. v. Meta Platforms, Inc., filed 2022-05-24), in which the '980 patent is one of the asserted patents per the FWD's mandatory notices. Separately, Unified Patents (a defensive aggregator) is actively attacking VL Collective's portfolio — but on other patents, e.g. IPR2022-01086 on U.S. 8,605,794, where the PTAB canceled claims 1, 3, 5, 9, 12–13, 15 and the Federal Circuit affirmed on 2026-02-20 (No. 24-1890). VL Collective has shown it does appeal adverse PTAB outcomes (24-1890; and it filed 25-1415 here), but here it withdrew rather than fight for the '980 patent. Finally, the patent's adjusted expiration is 2027-04-14 — roughly six-and-a-half months from today — which materially reduces the damages window.

Recommended next steps

1. Get the FWD and the certificate as primary sources before acting. The FWD is public: IPR2023-00924 Final Written Decision (patentlyo.com mirror). Pull the verbatim disposition from the FWD's conclusion and confirm it against the PTAB E2E / PTAB API docket and the IPR certificate (K1, issued 2025-12-19) to lock down exactly which claims the certificate cancels. The FWD states verbatim: "claims 1–3, 5–7, 9–11, and 13–15 are unpatentable, and Petitioner has not shown by a preponderance of evidence that claims 4, 8, 12, and 16 are unpatentable." Quote that in any Rule 11 letter.
2. If VL asserts claims 1, 2, 3, 5–7, 9–11, or 13–15, call it out immediately — those claims are canceled and cannot support infringement. Reference the FWD and the K1 certificate. If VL asserts claims 4, 8, 12, 16, demand its basis for enforceability given the cancellation of parent claims 1 and 9, and reserve the orphan-claim argument.
3. No pending PTAB trial to track. IPR2023-00924 is terminated; there is no active AIA trial on the '980 patent. So the "statutory 1-year deadline from institution / oral hearing / FWD due date" milestones do not apply — the only remaining clock is the 2027-04-14 expiration. If a new petitioner wants to attack claims 4/8/12/16, note that any such IPR would now face the PTO's discretionary-denial climate for older patents ("settled expectations") and parallel-litigation factors — a real but non-trivial hurdle.
4. Address the appellate record correctly. The only appeals tied to this patent are 25-1415 (dismissed 2025-08-11) and 25-1454 (dismissed 2025-10-14), both voluntary and both without any merits ruling. Do not cite the 2026 VL Collective v. Unified Patents decision (Fed. Cir. 24-1890, 2026-02-20) as precedent about this patent — it concerns U.S. 8,605,794 / IPR2022-01086, not 7,436,980.

Caveats / do-not-overstate: (a) I could not independently retrieve the certificate's verbatim claim listing — treat the cancel/survive table as derived from the FWD + the patent's own legal-events record, and confirm on PTAB E2E. (b) The dissent's author is unconfirmed from my excerpt. (c) The FWD is non-precedential and unreviewed, so the Board's broad construction of "spatio-temporal model" binds only that proceeding. (d) Google Patents' "Active, expires 2027-04-14" status predates the December 2025 certificate and should be treated as stale on the claim-status question.

I'll research the assignment records and the entities in the chain before compiling the analysis.

Let me verify the litigation vehicle and the recording correspondents.

I have enough to reconstruct the chain. Note: the Google Patents legal-events text and the earlier-generated summary contain owner/reel/frame/date data but no correspondent fields; I retrieved one correspondent (John S. Artz / Dickinson Wright PLLC) from the USPTO legacy-assignment PDF for reel 052142/0403. I flag the remaining gaps rather than invent them.

Inventors

All three executed the original assignment to Siemens Corporate Research, Inc. with signing dates spanning 2005-07-20 to 2005-09-09 (reel 016521/0376, recorded 2005-09-12).

Unusual-pattern note: There is no evidence of the "all inventors depart within 12 months of filing" fire-sale tell here. The inventors remained Siemens-affiliated; the patent stayed inside the Siemens corporate chain for ~14 years (2005 → 2019) before divestiture. This is a mature-portfolio monetization sale, not an early-abandonment. (I cannot confirm each inventor's exact tenure from the assignment record alone — that is an HR/SEC-level fact, not a recorded one.)

Original assignee

Siemens Corporate Research, Inc. (New Jersey), the U.S. corporate-R&D arm of Siemens AG — principal place of business Princeton, NJ.

• Primary line of business: central research lab (computer vision, medical imaging, image analytics). The patent's subject matter (boosted-detector object detection/tracking for vehicles and pedestrians) maps to Siemens' driver-assistance and video-surveillance interests.
• Product embodiment: Not established. This is a research-lab patent; I found no evidence that Siemens shipped a product practicing claims 1–16. The '980 patent was monetized through a portfolio sale, not productized.
• Current status of the original assignee: Siemens AG remains an operating company. The specific U.S. research entity was merged away (see timeline) and the asset was ultimately sold to VideoLabs, Inc. on 2019-09-25. The issuing assignee of record is no longer the owner.

Assignment timeline

Records below are from the Google Patents legal-events text supplied as authoritative; correspondent data is only available for one reel (sourced from the USPTO legacy-assignment PDF). Presented in execution/effective-date order.

• 2005-07-20 → 2005-09-09 (executed) / recorded 2005-09-12 — Reel 016521/0376

  • Conveyance: Assignment (inventors → company)
  • Assignor: Comaniciu, Dorin; Zhu, Ying; Sigal, Leonid
  • Assignee: Siemens Corporate Research, Inc. (New Jersey)
  • Correspondent: not captured in the supplied record (Google Patents legal events do not expose the correspondent field).
  • Context: standard inventor employment assignment at filing.

• 2009-09-02 (effective) / recorded 2010-04-05 — Reel 024185/0042

  • Conveyance: Merger
  • Assignor: Siemens Corporate Research, Inc.
  • Assignee: Siemens Corporation (New Jersey)
  • Correspondent: not captured in the supplied record.
  • Context: internal reorganization — SCR merged into Siemens Corporation.

• 2019-05-23 (effective) / recorded 2019-06-11 — Reel 049428/0966

  • Conveyance: Assignment
  • Assignor: Siemens Corporation
  • Assignee: Siemens Medical Solutions USA, Inc. (Pennsylvania)
  • Correspondent: not captured in the supplied record.
  • Context: internal portfolio re-papering within Siemens entities.

• 2019-06-18 (effective) / recorded 2019-06-19 — Reel 049511/0821

  • Conveyance: Assignment
  • Assignor: Siemens Medical Solutions USA, Inc.
  • Assignee: Siemens Healthcare GmbH (Germany)
  • Correspondent: not captured in the supplied record.
  • Context: internal portfolio re-papering, consolidating the asset under the German healthcare entity — the last Siemens owner before divestiture.

• 2019-09-25 (effective) / recorded 2022-07-13 — Reel 060500/0534

  • Conveyance: Assignment
  • Assignor: Siemens Healthcare GmbH
  • Assignee: VideoLabs, Inc. (California)
  • Correspondent: not captured in the supplied record.
  • Context: divestiture / transfer-to-licensing-entity — Siemens sold the asset to the VideoLabs patent-collective platform (Siemens Healthcare GmbH separately announced as a VideoLabs member effective 2019-11-01). Note the ~34-month recording lag (effective 2019-09-25, recorded 2022-07-13).

• 2019-12-04 (effective) / recorded 2020-02-07 — Reel 051748/0267

  • Conveyance: Security Interest (securitization)
  • Assignor/grantor: VL Collective IP LLC
  • Assignee/secured party: Praetor Fund I, a sub-fund of Praetorium Fund I ICAV (New York)
  • Correspondent: not captured in the supplied record.
  • Context: monetization financing — a litigation-finance-style security interest over the licensing entity's assets.

• 2019-12-19 (executed) / recorded 2020-03-17 — Reel 052142/0403

  • Conveyance: Assignment
  • Assignor: VideoLabs, Inc. (655 Bayport Avenue, San Carlos, CA 94070)
  • Assignee: VL IP Holdings LLC (California)
  • Correspondent: John S. Artz, Dickinson Wright PLLC (electronic-submission "Name of Submitter: JOHN S. ARTZ," signature /John S. Artz/; EPAS ID PAT6018684). Flag: this is the same correspondent firm/submitter of record on the paired down-chain LLC transfer — a recurrence, though Dickinson Wright is a large general-practice firm that does both operating-company and licensing work, so the single firm name is not by itself probative.
  • Context: shell/holding transfer — first hop of the VideoLabs → VL IP Holdings → VL Collective IP cascade (Attachment A lists 7,436,980 / 11/135,210 / US / 05/23/2005 / Graphical Object Models for Detection and Tracking).

• 2019-12-19 (executed) / recorded 2020-03-17 — Reel 052142/0508

  • Conveyance: Assignment
  • Assignor: VL IP Holdings LLC
  • Assignee: VL Collective IP LLC (California)
  • Correspondent: same recording event bundle as reel 052142/0403 (Dickinson Wright PLLC); confirm on the Assignment Center.
  • Context: shell-entity cascade — second hop, executed the same day as the upstream transfer, landing the asset in the active assertion vehicle.

• 2020-03-24 (effective) / recorded 2020-03-31 — Reel 052272/0435

  • Conveyance: Security Interest
  • Grantor: VL Collective IP LLC
  • Secured party: Praetor Fund I, a sub-fund of Praetorium Fund I ICAV
  • Correspondent: not captured in the supplied record.
  • Context: second monetization financing lien on the assertion vehicle.

• 2022-12-28 (effective) / recorded 2023-03-14 — Reel 062977/0325

  • Conveyance: Release by Secured Party
  • Assignor: Praetor Fund I, a sub-fund of Praetorium Fund I ICAV
  • Assignee: VL Collective IP LLC
  • Correspondent: not captured in the supplied record.
  • Context: lien release — the Praetor financing secured interests were discharged (consistent with a completed financing round/settlement).

Non-assignment events (for context, from the same legal-events text): IPR2023-00924 instituted at trial 2023-05-24 (petitioner: Meta Platforms, Instagram, WhatsApp, Meta Platforms Technologies, GIPHY); IPR certificate K1 issued 2025-12-19. Government-interest/entity-status events (small → big entity, 2023-09-26 → 2023-11-09) bracket the IPR.

Timeline diagram

timeline
    title Ownership of US 7436980
    2004 : Provisional filed
    2005 : App filed by Siemens Corporate Research
        : Inventors assign to Siemens
    2008 : Patent issued to Siemens
    2009 : Merger into Siemens Corporation
    2019 : Reassigned to Siemens Medical Solutions
        : Reassigned to Siemens Healthcare GmbH
        : Sold to VideoLabs Inc
        : Praetor Fund security interest
        : Chained transfer to VL IP Holdings
        : Chained transfer to VL Collective IP
    2020 : Recordings land at USPTO
        : Second Praetor security interest
    2022 : Suit filed against Meta in Delaware
    2023 : Praetor liens released
        : Meta IPR filed
    2025 : IPR certificate issued

NPE / troll-pattern signals

1. Shell-entity transfer — PRESENT.
The asset moved from an operating company (Siemens Healthcare GmbH) into a licensing-only structure via two same-day chained transfers: reel 052142/0403 (VideoLabs, Inc. → VL IP Holdings LLC) and reel 052142/0508 (VL IP Holdings LLC → VL Collective IP LLC), both executed 2019-12-19. "VL Collective IP" is a "wholly owned subsidiary of VideoLabs" per VideoLabs' own site; VideoLabs is a ~9-employee licensing platform headquartered in Palo Alto (address of record on the assignment: 655 Bayport Avenue, San Carlos, CA). No manufacturing.

2. Known asserter in the chain — PRESENT.
VideoLabs, Inc. / VL Collective IP LLC is a high-frequency patent plaintiff (D. Del., W.D. Tex. Waco, E.D. Tex.) suing Netflix, Roku, Amazon, Meta, Hisense, Microsoft and others. Taiwan's IP office (TIPO) characterizes VideoLabs as "廣義上屬 NPEs 之一種類型" (broadly a type of NPE). Directly on point: VideoLabs, Inc. v. Meta Platforms, Inc., No. 1:22-cv-00680 (D. Del., filed 2022-05-24) — the '980 patent is Count V, with "VL IP [Collective] … the assignee and lawful owner" (i.e., the reel 052142/0508 assignee).

3. Repeat correspondent across the chain — PARTIAL / PRESENT (on 2 links).
John S. Artz, Dickinson Wright PLLC is the submitting correspondent on reel 052142/0403 and the paired recording bundle for 052142/0508 — i.e., the two consecutive LLC-cascade links share one recorder, while the name of the licensed shell changes. Caveat: I could not retrieve the correspondents for reels 016521/0376, 024185/0042, 049428/0966, 049511/0821, and 060500/0534 from the sources available, so I cannot state whether Artz recurs beyond these two. A single firm appearing on a related pair is suggestive, not conclusive.

4. Cascading transfers — PRESENT.
Two distinct cascades: (a) Siemens side — reels 049428/0966 (eff. 2019-05-23), 049511/0821 (eff. 2019-06-18), 060500/0534 (eff. 2019-09-25) within ~4 months; (b) VideoLabs side — reels 052142/0403 and 052142/0508, both executed 2019-12-19, i.e., chained LLC hops on a single day.

5. Pre-litigation transfer — NOT PRESENT (for this patent's first suit).
The transfer into the assertion vehicles occurred in 2019; the first suit asserting the '980 patent was filed 2022-05-24 (1:22-cv-00680). That is ~29 months, well outside the 6-month window. (Note, however, that VideoLabs' complaint states it began contacting Meta "at least as early as October 2, 2019" — i.e., pre-suit licensing outreach tracked the acquisition, even though the suit did not.)

6. Bankruptcy fire-sale — NOT PRESENT.
No Chapter 7/11 anywhere in the chain. Siemens AG was solvent; the '980 patent left the corporate family in an ordinary discretionary portfolio divestiture (reel 060500/0534).

7. Privateering — UNCLEAR.
Siemens sold the asset to VideoLabs and simultaneously joined VideoLabs as a member (Siemens Healthcare GmbH membership effective 2019-11-01), and VideoLabs is "industry-sponsored" (Soryn IP Group / Bardin Hill / Halcyon backing). That is a collective-licensing structure rather than classic competitor-financed privateering — Siemens does not appear to direct assertions against competitors. Insufficient evidence to call it privateering.

8. Defensive aggregator (anti-NPE) — NOT PRESENT.
The chain does not terminate at RPX / AST / LOT / OIN. VideoLabs markets itself as a collective but asserts as a plaintiff (Netflix, Roku, Amazon, Meta, Hisense, Microsoft) and the portfolio has taken repeated IPR losses — the opposite of a neutralizing endpoint.

Verdict

NPE — high confidence.

The record shows (a) a transfer from operating owner Siemens Healthcare GmbH into a two-hop licensing structure — reel 060500/0534 (eff. 2019-09-25) → reel 052142/0403 → reel 052142/0508 — landing in VL Collective IP LLC, a wholly owned VideoLabs subsidiary; (b) an identified high-frequency asserter that litigated this exact patent against Meta in 1:22-cv-00680 (D. Del., filed 2022-05-24), triggering IPR2023-00924 (Meta et al.); and (c) litigation-finance-style securitization via two Praetor Fund I security interests (reels 051748/0267 and 052272/0435), later released (reel 062977/0325). Those are three independent, evidence-backed NPE signals with the shell-transfer, known-asserter, and cascade elements all anchored to specific reels and dates.

Verification link: USPTO Assignment Center search for patent 7,436,980 — https://assignmentcenter.uspto.gov/ (mirror: https://assignment.uspto.gov/patent/index.html). Corroborating recorded instrument for reel 052142/0403: http://legacy-assignments.uspto.gov/assignments/assignment-pat-052142-0403.pdf

Two refinements to the earlier-generated section, for consistency:

1. The prior summary could not confirm the Feb 20, 2026 Federal Circuit item. It is now resolved: the CAFC opinion of 2026-02-20 is Appeal No. 24-1890, VideoLabs v. Unified Patents, concerning U.S. 8,605,794 (the '794 patent) — not 7,436,980. So the 2026 docket item does not concern this patent; the only Federal Circuit matters tied to '980 remain the dismissed 2025 appeals (25-1415, 25-1454).
2. The prior summary listed "current listed assignees VL Collective IP LLC and VL IP Holdings LLC." The IPR2023-00924 FWD's real-parties-in-interest statement confirms the corporate nesting: VL Collective ← VL IP Holdings LLC (parent) ← VideoLabs, Inc. (parent). That matches the reel 052142/0403 → /0508 chain.

Unverified items I am explicitly NOT asserting: the correspondents of record for the Siemens-era reels (016521/0376, 024185/0042, 049428/0966, 049511/0821, 060500/0534) and for the Praetor reels; and whether John S. Artz recurs on any other recordings. Retrieve these directly from the Assignment Center before relying on them.

One naming inconsistency worth flagging: the body of the recorded instrument at reel 052142/0403 reads "Assignment From VideoLabs, Inc. to VLP Holdings LLC" while the cover sheet and Google Patents legal events name the assignee "VL IP Holdings LLC." Same reel/frame, likely a typo in the instrument — note it if prosecuting or clearing title.

I'll verify this specific patent and gather the prior-art citations. Let me search the USPTO/patent databases.

Let me verify the individual prior-art references cited on the face of this patent.

Let me gather details on the non-patent literature cited and the family-cited references, plus the IPR art.

Note: the follow-up searches I attempted exceeded the tool-step limit, so the NPL/family details below rely on the authoritative full patent text you supplied plus the three searches already returned. I flag anything unverified.

Prior-Art Analysis — US 7,436,980 B2 ("Graphical object models for detection and tracking")

0. Reference identification (literal, no auto-correction)

The number 7,436,980 resolves to US 7,436,980 B2, App. No. 11/135,210, filed 2005-05-23, priority 2004-05-27 (Prov. 60/574,799), issued 2008-10-14, inventors Sigal / Zhu / Comaniciu, original assignee Siemens Corporate Research. This is consistent with the earlier section; I build on it rather than repeat it.

Important framing for the §102 question. The citations that appear on the face of this patent are a short list of background references. As explained in §5 below, none of the three "Patent Citations" discloses every element of independent claim 1 or claim 9, so none is a clean single-reference §102 anticipation. I state that explicitly rather than manufacture an anticipation match. The task's real-world "most relevant prior art" for this patent is the art that was actually run in the PTAB (the Zhao reference), which I cover in §6.

1. Patent citations appearing on the face of US 7,436,980 (3 references)

From the patent's "Patent Citations (3)" block:

Caveat on provenance: the Google Patents rendering marks US 7,200,266 with an asterisk (examiner-cited) and shows US 6,597,801 and EP 1 217 572 without that mark. I cannot fully confirm the examiner-vs-third-party attribution from the jumbled rendering, so I treat all three simply as references of record. All three antedate the 2004-05-27 priority date, so each is at least presumptively §102(b)/§102(e) art.

Reference A — US 6,597,801 B1 (Hewlett-Packard)

• Full citation: US 6,597,801 B1, "Method for object registration via selection of models with dynamically ordered features," Hewlett-Packard Development Company L.P.; filed/priority 1999-09-16; granted 2003-07-22. (A sibling is US 6,618,490 B1, same family.)
• Brief description: Registers a plurality of object models in an image one feature per model at a time, where the order in which features are searched is chosen dynamically to minimize search cost, and model states are updated according to intra-model and inter-model constraints (i.e., a constraint graph over model states sharing a common relationship). It is fundamentally a model-registration / tracking-efficiency invention.
• Relevance / §102 analysis:
  • Claim 3 / 11 (graphical model with nodes for components and the object): This is the only patent reference with a genuine structural echo — its "inter-model constraints" restricting model states to "a shared relationship" resemble the constraint edges of the claimed two-layer model. But it describes object models, not an explicit two-layer spatio-temporal graph with a coarse object node and component nodes, and no temporal-window factor.
  • Claim 2 / 10 (a detector per component): It searches for features of each model, but does not disclose an AdaBoost-style component detector in the claimed sense.
  • Claims 1 / 9 as a whole: No anticipation. It lacks the "provide a video of multiple images," the "measure the object as a collection of components in each image," the probability computation over a spatio-temporal model, and the threshold comparison. Best characterized as §103 background art (constraint-based multi-model registration).
• §102 posture: §102(b) art (publicly available well over one year before 2005-05-23). Potentially anticipatory only of the broad "constraint between model states" idea underlying claim 3; not of any full claim.

Reference B — EP 1 217 572 A2 (Eastman Kodak)

• Full citation: EP 1 217 572 A2 (A3 published 2004-01-14), "Digital image processing method and computer program product for detecting human irises in an image," Eastman Kodak Company (Shoupu Chen; Lawrence A. Ray); priority US 09/740,422 (2000-12-19); A2 published 2002-06-26. U.S. counterpart: US 6,920,237 B2 / US 2002/0106112 A1.
• Brief description: Detects irises by (i) measuring red pixel intensity, (ii) computing the probability a pixel is / is not an iris, and (iii) deciding pixel-by-pixel using a Bayes model and a probability-relationship test; an embodiment finds an oval skin-color region and locates irises within it.
• Relevance / §102 analysis:
  • Claims 1 / 9, element "determining a probability": Superficially touches the "probability" wording, but it is per-pixel classification, not a probability that an object as a collection of components is present in an image.
  • Claim 6 / 14 (position and size): It locates eye/iris position, but not object position-and-scale in a component model.
  • Claim 7 / 15 (empirical threshold): A decision threshold is implicit in a Bayes test, but the reference does not teach an empirically determined object-detection threshold.
  • Conclusion: No anticipation of any claim. It is remote from component-based, spatio-temporal object modeling. The examiner almost certainly cited it only as illustrative of probabilistic (Bayes) detection reasoning. Best viewed as §103-adjacent background only.
• §102 posture: §102(b) art (published 2002-06-26, >1 yr before filing).

Reference C — US 7,200,266 B2 (Princeton University)

• Full citation: US 7,200,266 B2, "Method and apparatus for automated video activity analysis," Princeton University (I. Ozer; Tiehan Lu; Wayne Wolf); filed 2002-08-27; granted 2007-04-03.
• Brief description (from the published abstract/summary): Detects any new object introduced into a known environment, classifies object regions into human body parts (e.g., head, arms) or other rigid/non-rigid objects, compares detected objects against graphs from a database to identify parts and decide on presence of an object of interest, and tracks the movement of different object parts, later combining part-motion patterns by a classifier into a high-level activity.
• Relevance / §102 analysis — this is the most substantive of the three patent references:
  • Claim 1 / 9, elements "video of a plurality of images" + "measure the object as a collection of components in each image": Reference C discloses detecting and classifying object parts across video frames — a genuine overlap with the "collection of components" concept.
  • Claim 2 / 10 (detector for each component): It uses classifiers/detectors for object regions and body parts — close, though not the claimed per-component detector architecture.
  • Claim 3 / 11 (graphical model): It "compares … with the graphs from a database" and models parts — an overlap with a graph-based representation, but not the claimed two-layer spatio-temporal graphical object model.
  • Claim 5 / 13 (temporal window): It tracks movements over time, but does not disclose a temporal window of width w passing component measurements into the current image.
  • Claim 8 / 16 (specific joint probability distribution ψ/φ factorization): Not disclosed. This is the patent's signature limitation and no reference here approaches it.
  • Conclusion: No clean anticipation of claim 1 or 9 as a whole — it lacks the spatio-temporal graphical model, belief-propagation inference, and the thresholded object-presence decision. It is, however, the strongest single-reference combination candidate (component detection + graph comparison + temporal part tracking) and is best paired under §103 with a belief-propagation reference such as Isard (PAMPAS) or Sudderth.
• §102 posture: §102(e) art — its 2002-08-27 filing predates the 2004-05-27 priority date, so it qualifies as prior art as of its filing date even though it did not publish until 2007.

2. Family-cited references (cited in the family members, not on the U.S. front page)

These appear under "Family Cites Families (6)" — i.e., they were cited in the EP/JP/DE counterparts. They carry earlier publication dates and are therefore also §102(b)-eligible, but again none reaches the spatio-temporal-model + component-belief-propagation combination.

§102 conclusion for the family cites: background only. None discloses a two-layer spatio-temporal graphical object model, message passing, or the claim-8/16 joint distribution.

3. Non-patent literature cited on the face of the patent (8 references)

These are the references most likely to have driven any §103 rejection, because they describe the exact building blocks the patent combines.

§102 (§102(a)/(b)) analysis of the NPL:

• References 1, 2, 6, 7, 8 are clearly pre-2004 §102(b) art.
• References 3, 4, 5 are 2004 conference papers. If published after the 2004-05-27 provisional priority date (Lan et al. and Sigal et al. were presented at CVPR June 27-Jul 2, 2004; Okuma at ECCV 2004), they are not §102(b) art and would only be §102(a) art if shown to predate the applicant's actual date of invention. This is a genuine timing flag — treat their §102 status as unresolved.
• Anticipation of claim 8 / 16 (the specific ψ/φ joint distribution): none of the NPL discloses the exact claim-8/16 factorization; that is why claims 8 and 16 survived the IPR (see §6).

4. Bottom line on the face-of-patent citations

• No single cited reference anticipates independent claim 1 or claim 9. Each cited reference supplies at most one or two of the five recited steps.
• §102 is essentially inapplicable to these citations; the examiner's use of them is best understood as §103 (individually or in combination) or as background.
• Strongest single face reference: US 7,200,266 B2 (component/part detection + graph comparison + temporal part tracking).
• Strongest combination story under §103: US 7,200,266 (or Lan et al.) for the component/spatio-temporal model + Isard PAMPAS / Sudderth (non-parametric belief propagation) + Viola-Jones (boosted component detectors).

5. The art that actually matters (from the IPR) — Zhao

The prior art the PTAB actually relied on in IPR2023-00924 (Meta et al. v. VL IP Holdings) was the Zhao reference — "Dressed Human Modeling, Detection, and Parts Localization" — a parts/component-based human detection reference, per the earlier section of this analysis and the PTAB petition materials surfaced in search.

• The Board's Final Written Decision (per the earlier section) found claims 1-3, 5-7, 9-11, 13-15 unpatentable, and left claims 4, 8, 12, 16 not shown unpatentable (the prior+later-image measurement limitation and the specific ψ/φ joint-distribution limitation).
• I could not re-verify this ground chart with a fresh search this turn (tool step limit), so treat the Zhao characterization as carried over from the prior section and the PTAB petition PDF rather than independently reconfirmed here.

Contradiction to flag: the earlier summary gives the FWD date as "Dec. 5, 2024," whereas the authoritative full text you supplied shows the IPR certificate (K1) issued 2025-12-19 (effective 20251219). The two are in tension (an FWD in Dec 2024 could ripen into a Dec 2025 certificate, so they are reconcilable, but the earlier summary's "FWD Dec. 5, 2024" is not confirmed by the supplied text and should be checked against the certificate/PTAB record).

6. Caveats and open items

1. No citation here is a §102 anticipation of claims 1 or 9. I state this deliberately; the patent's breadth is narrow enough that only a combination theory works, and the only limitation that consistently survived challenge is claim 8/16's specific joint distribution.
2. §102(e) reference C (US 7,200,266) qualifies as prior art via its 2002-08-27 filing date, not its 2007 grant date.
3. The three 2004 conference papers (Okuma, Sigal, Lan) may post-date the provisional priority date — their §102 status is unresolved and should be checked against the actual publication dates.
4. Examiner-vs-third-party attribution of the three patent citations is ambiguous in the Google Patents rendering.
5. Unverified this turn: the Zhao reference's full bibliographic record and the IPR ground chart; I did not retrieve them because the search tool reached its step limit. Verify against the IPR2023-00924 petition and Final Written Decision.

Sources: Google Patents US7436980B2 (authoritative full text supplied, including "Patent Citations (3)," "Family Cites Families (6)," "Non-Patent Citations (8)," and "Citations (3)"); Google Patents US6597801B1; Google Patents EP1217572A3 / EPO publication server EP 1 217 572 A3 (filing 2001-12-06, A2 pub. 2002-06-26); Google Patents US7200266B2 (2002-08-27 / 2007-04-03).

I'll ground this analysis against the prior-art record before drafting the §103 combinations.

Obviousness Analysis — U.S. Patent 7,436,980 B2 under 35 U.S.C. § 103

Note on framework: The application was filed May 23, 2005 (priority May 27, 2004), so pre‑AIA § 103(a) governs. This analysis builds on the previously generated summary and does not repeat the claim text, family, or assignment history already given there.

I. The hypothetical POSITA

A person having ordinary skill at the May 2004 priority date would hold an M.S. or Ph.D. in computer vision/electrical engineering (or equivalent), with 2–4 years of experience in object detection and tracking, and would be familiar with: (a) boosted/AdaBoost part detectors (Viola–Jones); (b) probabilistic graphical models and message passing/belief propagation; (c) particle filtering and its variants; and (d) articulated-object part models in video. Critically, all of the references below were authored by, or are contemporaneous with, the inventors' own research community (Isard, Viola, Okuma, Sigal, Lan/Huttenlocher) — this is a mature, well-populated art where combination is the norm.

II. Claim 1 / claim 9 limitation breakdown

Dependents add: detectors per component (2/10); two‑layer graphical model (3/11); prior-image + later-image measurements (4/12); temporal window (5/13); position and size (6/14); empirical threshold (7/15); the specific joint-probability factorization (8/16).

III. Prior art inventory (the "Prior Art" section of the page)

Patent references cited on the face:

• US 7,200,266 B2 (Princeton Univ.) — "Method and apparatus for automated video activity analysis." (Referred to in the IPR record as "Ozer" — see §IV‑C.)
• US 6,597,801 B1 (Hewlett-Packard) — object registration via dynamically ordered features.
• EP 1 217 572 A2 (Eastman Kodak) — detecting human irises via component/feature classification.

Non-patent references cited on the face:

• Viola & Jones (CVPR 2001) — rapid object detection with a boosted cascade of simple features.
• Viola, Jones & Snow (ICCV 2003) — detecting pedestrians using patterns of motion and appearance.
• Isard, "PAMPAS" (CVPR 2003 / 2004) — real‑valued graphical models; non‑parametric belief propagation.
• Sudderth, Ihler, Freeman & Willsky (MIT LIDS TR P‑2551, Oct. 15, 2002) — non‑parametric belief propagation.
• Okuma, Taleghani, de Freitas, Little & Lowe (ECCV 2004) — "A Boosted Particle Filter: Multitarget Detection and Tracking."
• Sigal, Bhatia, Roth, Black & Isard (CVPR 2004) — "Tracking Loose‑limbed People" ("TLP" in the IPR).
• Lan & Huttenlocher (CVPR 2004) — "A Unified Spatio‑Temporal Articulated Model for Tracking."

References also in the family/prior‑art record used in the IPR:

• Zhao, "Dressed Human Modeling, Detection, and Parts Localization," CMU‑RI‑TR‑01‑19 (2001) (doctoral dissertation).
• Steffens, U.S. 6,301,370 (vision‑based object detection/tracking with a node‑based graphical model).

Grounding: the Google Patents record for US7436980B2 lists US7200266B2, US6597801B1, EP1217572A2 and the eight NPL items above. The IPR2023‑00924 Final Written Decision (FWD) identifies Zhao, Steffens, Ozer (= US 7,200,266) and TLP as the Petitioner's grounds, and confirms the claims held unpatentable vs. not shown unpatentable.

IV. Combinations that render the claims obvious

Ground A — Zhao alone (single-reference § 103)

This is the strongest, already-adjudicated ground. Zhao discloses an "integrated human shape modeling, detection, and body part localization" system that decomposes a segmented silhouette into body parts (torso/limbs — i.e., 1(c)), encodes spatial relationships among parts in a human model, and incorporates temporal motion information from previous frames to disambiguate limb orientation (i.e., 1(a)). Zhao's Bayesian Similarity Measure ("BSM") scores the resemblance of a contour to the model, and Zhao expressly detects a person where BSM(C) ≥ threshold (i.e., 1(d) and 1(e)).

The PTAB (Majority) found Zhao rendered claims 1–3, 5–7, 9–11, 13–15 unpatentable. That is a powerful, evidence‑backed § 103 conclusion for the independent claims and the detector / graphical‑model / window / position‑and‑size / empirical‑threshold dependents. The Board reasoned that Zhao's motion information is derived from objects detected by comparing a likelihood to a threshold in prior frames, and is used when comparing the current frame to the threshold. (FWD at 9, 4.)

Ground B — Steffens + Zhao

Steffens (U.S. 6,301,370) discloses a node/edge graphical model of an object plus tracking of those nodes across frames — supplying the two‑layer graphical‑model element of claims 3/11 explicitly. Steffens uses an undefined "similarity" measure; Zhao supplies the probabilistic implementation (BSM). Motivation: both address the same problem with structurally similar spatio‑temporal node models; a POSITA implementing Steffens with a defined likelihood would look to Zhao's Bayesian framework → KSR rationale (d): applying a known technique to a known device ready for improvement. (See the ex‑parte summary of Grounds 2–3.)

Ground C — Ozer (US 7,200,266 B2) + Zhao

US 7,200,266 B2, cited on the patent's own face as examiner‑cited art, discloses automated video activity analysis over time — i.e., a temporal/spatio‑temporal modeling component over image sequences. Combined with Zhao (or with Viola–Jones for the part detectors), it supplies the temporal‑window / sequence‑reasoning element (claims 5/13).

Ground D — TLP (Sigal et al., CVPR 2004)

TLP was cited on the IDS but never applied by the Examiner. It expressly teaches a probabilistic graphical model with nodes for body parts and edges encoding spatial and temporal relationships (claims 1(a), 3/11), combined with a probabilistic image‑likelihood model (claim 1(d)); belief estimation in the graph yields the detection decision (claim 1(e)). Under KSR, a POSITA would reason that estimating belief and thresholding it is the natural, predictable completion of the inference step.

Ground E — The prosecution‑cited combination: Viola–Jones + Okuma boosted particle filter + Isard PAMPAS/Sudderth NBP + Lan & Huttenlocher

This is the combination the Examiner apparently did not squarely address, and it maps every element, including claims 4/8/12/16, which survived the IPR:

Why this combination is legally compelling:

• 1(a), 1(c), 3/11 are met by Lan/Huttenlocher (two‑dimensional spatio‑temporal part model; parts with spring constraints) and by TLP.
• 1(d)–1(e) are met by Viola–Jones (part classifiers + threshold) and Okuma (detection hypotheses feeding a probabilistic tracker).
• 2/10, 6/14, 7/15 are squarely met by Viola–Jones (per‑part boosted detectors at multiple scales with empirically tuned thresholds).
• 4/12, 5/13 (prior AND later image / temporal window) are met by Lan & Huttenlocher's HMM temporal model (which natively supports forward–backward / fixed‑lag smoothing) and by Isard PAMPAS, whose whole premise is belief propagation producing a smoothed posterior from messages on both sides of a node.
• 8/16 (the joint‑distribution formula) — see §VII.

V. Claim‑by‑claim chart (Ground E)

VI. Motivation to combine (KSR rationales, all applicable)

1. Same field, same problem, same solution type. Every reference targets object/part detection and tracking in video. KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398, 417 (2007).
2. Predictable result. Substituting Zhao's likelihood or Viola–Jones's detector into a Lan/Huttenlocher or Isard graphical model yields nothing more than the expected aggregation of part evidence — exactly what the patent's own "Simple Fusion" comparison shows fails (see §VIII).
3. Known technique improving a similar device (Okuma: AdaBoost proposals improving a particle filter).
4. Design incentive / finite predictable options. For the smoothing window there are only two known choices — detection‑centric re‑initialization or tracking‑centric propagation — and the specification itself concedes both.
5. Reasonable expectation of success, evidenced by the references' own citations to one another (Okuma cites Viola–Jones; Lan cites Felzenszwalb/Huttenlocher pictorial structures; TLP and PAMPAS share authors with the patent).

VII. The claims that survived the IPR (4, 8, 12, 16) — how to reach them

The FWD held Petitioner had not shown claims 4, 8, 12, 16 unpatentable. The reason was evidentiary, not conceptual:

Claim 4 / 12. The Board rejected the theory that Zhao's "motion information" teaches "comparing the likelihood of object presence in each image to a threshold"; the only record testimony (Dr. Saber) refuted that theory. But claim 4's actual text ("determined from a prior image and a later image relative to the current image") is precisely a fixed‑interval smoother, and Lan & Huttenlocher's HMM temporal model and Isard's PAMPAS disclose using messages/frames on both sides of the current node. A challenger using those references — rather than Zhao's sparse motion‑information sentence — would have a materially stronger record. (The Board also flagged a live claim‑construction dispute: whether 1(e) requires comparing multiple probabilities from multiple images.)

Claim 5 / 13 and 7 / 15 already fell with claim 1 under Ground A; the windowed‑smoothing and empirical‑threshold limitations did not save them.

Claim 8 / 16 (the joint‑probability formula). This is a textbook chain‑rule factorization of a directed (Bayesian‑network) graphical model with the conditional independencies the patent itself asserts (conditional independence of components over time given the object appearance). Given the two‑layer model of claim 3 and the stated independence assumption, the product of object–object temporal potentials ψᵢⱼ(Xᵢᴼ,Xⱼᴼ), object–component spatial potentials ψᵢₖ(Xᵢᴼ,Xᵢ^Ck), component–component potentials ψₖₗ(Xᵢ^Ck,Xᵢ^Cl), and local‑evidence terms φᵢ follows directly from standard graphical‑model semantics (Pearl, Probabilistic Reasoning in Intelligent Systems; Koller & Friedman, Probabilistic Graphical Models — both pre‑2004). The § 103 theory is "obvious to try" / routine mathematical expression of a known model class; its survival at the PTAB reflects that Petitioner did not put in the evidentiary proof, not that the formula is inventive.

VIII. Secondary considerations (Graham) and anticipated rebuttals

• Unexpected results — weak. The patent's only data ("Tests") compares the full model against a deliberately naive "Simple Fusion" that averages each component's best detection. That is a comparison to an admitted baseline, not to the claimed combination of the prior art; there is no nexus showing the claimed combination outperforms Zheng/Viola–Jones/Lan. Nor is this evidence of non‑obviousness in the litigation record.
• Long‑felt need / industry adoption — not in the intrinsic record. No evidence of copying or licensing tied to the claims surfaced; the patent's value appears to reside in the portfolio/litigation context (IPR + Delaware 1:22‑cv‑00680 + CAFC 25‑1415/25‑1454), not in commercial embodiment.
• Teaching away — none. No reference criticizes combining component detectors with graphical‑model inference; to the contrary, Okuma and TLP affirmatively advocate it.

IX. Bottom line and confidence

The single most defensible proposition: claims 1–3, 5–7, 9–11, 13–15 are invalid over Zhao, as the PTAB has already found. The contested frontier is 4/8/12/16, and the winning theory is not Zhao's weak "motion information" sentence but Lan & Huttenlocher's forward/backward HMM (claim 4) and the standard Bayesian‑network chain‑rule factorization (claim 8).

Caveats / verify before relying:

1. The claim‑construction dispute over "comparing the probabilities for each image to a threshold" is unresolved and outcome‑determinative; under a narrower reading (multiple probabilities, multiple images), the Zhao‑alone ground for 1(e) weakens.
2. I have not retrieved the full Steffens U.S. 6,301,370 text, the Ozer US 7,200,266 disclosure, or Dr. Saber/Bajaj expert testimony firsthand; the Ozer↔US 7,200,266 and TLP↔Sigal 2004 equivalences are inferences from the ex‑parte IPR summary, though consistent with US 7,200,266 appearing as examiner‑cited art and Sigal 2004 appearing in the IDS.
3. The IPR certificate (K1) issued Dec. 19, 2025 cancels claims 1 and 9 while 4/8/12/16 (which depend from them) were not shown unpatentable. Confirm the exact enforceable claim set against the USPTO certificate before relying on any single claim's validity.
4. The effect of the 2023‑24 post‑grant re‑assignment to VideoLabs/VL Collective IP on standing to assert surviving claims is outside the § 103 question.

Sources: Google Patents US7436980B2 (authoritative full text); IPR2023‑00924 Final Written Decision, patentlyo.com/media/2025/04/IPR2023-00924-FD-ready-for-mailing.pdf; ex‑parte case summary, ai-lab.exparte.com/case/ptab/IPR2023-00924; Zhao thesis, ri.cmu.edu/pub_files/pub4/zhao_liang_2001_1/zhao_liang_2001_1.pdf; Okuma et al., cs.ox.ac.uk/publications/publication7522; Lan & Huttenlocher, ieeexplore.ieee.org/document/1315103 and semanticscholar.org/CorpusID:10366469.