IP Library Granted Patent US 12,561,777
Granted Patent B2
US 12,561,777 · App. 18/312,085 · Granted Feb 24, 2026

Anomaly detection using neural radiance fields

Inventors: Eliyahu Schwartz (Haifa, IL); Leonid Karlinsky (Acton, MA); Assaf Arbelle (Lehvot Haviva, IL); Sivan Harary (Manof, IL); Roi Herzig (Hod Hasharon, IL)
Assignee: International Business Machines Corporation
G06T7/0002G06T11/00G06T2207/10024G06T2207/10028G06T2207/20081G06T2207/20084G06T2207/20212G06T2207/30108
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,561,777
App. No.
18/312,085
Filed
May 4, 2023
Granted
Feb 24, 2026
Kind
B2
Art Unit
2675
USPC
382/156
Abstract

Detecting data anomalies by receiving a query image, determining a query image viewpoint according to a trained neural radiance field model, generating a 2D reference image according to the neural radiance field model, determining a difference between the query image and the reference image, and highlighting the difference in a presentation of the query image.

Claims (59)

1 . A computer implemented method for detecting data anomalies, the method comprising:

receiving a query image;

determining a query image viewpoint according to a trained model by minimizing rendering loss between weights of the trained model and the query image;

generating a two-dimensional reference image according to the query image viewpoint determined by the model;

determining a difference between the query image and the reference image; and

highlighting the difference in a presentation of the query image.

2 . The computer implemented method according to claim 1 , further comprising:

receiving a of set images;

training a neural radiance field model according to the set of images; and

providing the trained neural radiance field model.

3 . The computer implemented method according to claim 2 , further comprising freezing the weights of the neural radiance field model.

4 . The computer implemented method according to claim 1 , wherein determining the query image viewpoint comprises optimizing the neural radiance field model for the query image viewpoint.

5 . The computer implemented method according to claim 1 , wherein determining the difference between the query image and the reference image comprising comparing query image and reference image pixels.

6 . The computer implemented method according to claim 1 , further comprising

defining a query image depth map;

defining a reference image depth map;

determining a difference between the query image depth map and the reference image depth map; and

presenting details about the difference to a user.

7 . The computer implemented method according to claim 6 , further comprising determining the difference using a red-green-blue-depth camera.

8 . A computer program product for detecting anomalies, the computer program product comprising one or more computer readable storage media and collectively stored program instructions on the one or more computer readable storage media, the stored program instructions which, when executed, cause one or more computer systems to:

receive a query image;

determine a query image viewpoint according to a trained neural radiance field model by minimizing rendering loss between weights of the trained neural radiance field model and the query image;

generate a two-dimensional reference image according to the query image viewpoint determined by the model neural radiance field model;

determine a difference between the query image and the reference image; and

highlight the difference in a presentation of the query image.

9 . The computer program product according to claim 8 , the computer program instructions further causing the one or more computer systems to:

receive a set of images;

train a neural radiance field model according to the set of images; and

provide the trained neural radiance field model.

10 . The computer program product according to claim 9 , the computer program instructions further causing the one or more computer systems to freeze the weights of the neural radiance field model.

11 . The computer program product according to claim 8 , wherein determining the query image viewpoint comprises optimizing the neural radiance field model for the query image viewpoint.

12 . The computer program product according to claim 8 , wherein determining the difference between the query image and the reference image comprising comparing query image and reference image pixels.

13 . The computer program product according to claim 8 , the computer program instructions further causing the one or more computer systems to:

define a query image depth map;

define a reference image depth map;

determine a difference between the query image depth map and the reference image depth map; and

present details about the difference to a user.

14 . The computer program product according to claim 13 , further comprising determining the difference using a red-green-blue-depth camera.

15 . A computer system for detecting data anomalies, the computer system comprising:

one or more computer processors;

one or more computer readable storage devices; and

stored program instructions on the one or more computer readable storage devices for execution by the one or more computer processors, the stored program instructions which, when executed, cause the one or more computer processors to:

receive a query image;

determine a query image viewpoint according to a trained neural radiance field model by minimizing rendering loss between weights of the trained neural radiance field model and the query image;

generate a two-dimensional reference image according to the query image viewpoint determined by the model neural radiance field model;

determine a difference between the query image and the reference image; and

highlight the difference in a presentation of the query image.

16 . The computer system according to claim 15 , the computer program instructions further causing the one or more processors to:

receive a set of images;

train a neural radiance field model according to the set of images; and

provide the trained neural radiance field model.

17 . The computer system according to claim 16 , the computer program instructions further causing the one or more processors to freeze the weights of the neural radiance field model.

18 . The computer system according to claim 15 , wherein determining the query image viewpoint comprises optimizing the neural radiance field model for the query image viewpoint.

19 . The computer system according to claim 15 , wherein determining the difference between the query image and the reference image comprising comparing query image and reference image pixels.

20 . The computer system according to claim 15 , the computer program instructions further causing the one or more processors to:

define a query image depth map;

define a reference image depth map;

determine a difference between the query image depth map and the reference image depth map; and

present details about the difference to a user.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded May 4, 2023
From: SCHWARTZ, ELIYAHU; KARLINSKY, LEONID; ARBELLE, ASSAF; HARARY, SIVAN; HERZIG, ROI
To: INTERNATIONAL BUSINESS MACHINES CORPORATION
Reel/Frame 063535/0010 →
Continuity (1)
Related Publication 20240370983A1 · Nov 7, 2024
References Cited (19)
US 11172148B2 · Overbeck · 2021 [cited by examiner]
US 11308675B2 · Wang · 2022 [cited by examiner]
US 11557027B2 · Wen · 2023 [cited by examiner]
US 12073507B2 · Xu · 2024 [cited by examiner]
US 12223594B2 · Coffman · 2025 [cited by examiner]
US 20020039436A1 · Alumot · 2002 [cited by examiner]
US 20150265187A1 · Bernal · 2015 [cited by examiner]
US 20240273811A1 · Radwan · 2024 [cited by examiner]
NO 2022119844A1 · 2022 [cited by applicant]
Tancik, et al. (Block-NeRF: Scalable Large Scene Neural View Synthesis) (Year: 2022). [cited by examiner]
Ehret, et al., “Image Anomalies: a Review and Synthesis of Detection Methods,” Journal of Mathematical Imagng and Vision Manuscript, arXiv:1808.02564v2 [cs.CV] Jun. 3, 2019, 36 Pgs. [cited by applicant]
Jeong, et al., “Time-Series Anomaly Detection with Implicit Neural Representation,” arXiv:2201.11950v1 [cs.LG] Jan. 28, 2022, 9 Pgs. [cited by applicant]
Lee, et al., “AnoViT: Unsupervised Anomaly Detection and Localization with Vision Transformer-based Encoder-Decoder,” arXiv:2203.10808v1 [cs.CV] Mar. 21, 2022, 9 Pgs. [cited by applicant]
Marimont, et al., “Implicit field learning for unsupervised anomaly detection in medical images,” arXiv:2106.05214v1 [eess.IV] Jun. 9, 2021, 10 Pgs. [cited by applicant]
Martin-Brualla, et al., “NeRF in the Wild: Neural Radiance Fields for Unconstrained Photo Collections,” arXiv:2008.02268v3 [cs.CV] Jan. 6, 2021, 15 Pgs. [cited by applicant]
Mildenhall, et al., “NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis,” arXiv:2003.08934v2 [cs.CV] Aug. 3, 2020, 25 Pgs. [cited by applicant]
Rematas, et al., “Novel Views of Objects from a Single Image,” arXiv:1602.00328v2 [cs.CV] Aug. 15, 2016, 14 Pgs, <https://arxiv.org/abs/1602.00328#:˜:text=Taking%20an%20image%20of%20an,final%20and%20not%20easily%20rever… [cited by applicant]
Wang, et al., “NeRF—: Neural Radiance Fields Without Known Camera Parameters,” arXiv:2102.07064v4 [cs.CV] Apr. 6, 2022, 17 Pgs. [cited by applicant]
Yu, et al., “pixelNeRF: Neural Radiance Fields from One or Few Images,” Accessed on Feb. 5, 2023, 10 Pgs, <https://openaccess.thecvf.com/content/CVPR2021/papers/Yu_pixelNeRF_Neural_Radiance_Fields_From_One_or_Few_Images… [cited by applicant]