US 12,406,477 B2
Iterative refinement of annotated datasets
Willem Verbeke, Gothenburg (SE)
Assigned to ZENSEACT AB, Gothenburg (SE)
Filed by ZENSEACT AB, Gothenburg (SE)
Filed on Feb. 6, 2023, as Appl. No. 18/164,808.
Claims priority of application No. 22156360 (EP), filed on Feb. 11, 2022.
Prior Publication US 2023/0260257 A1, Aug. 17, 2023
Int. Cl. G06V 10/00 (2022.01); G06V 10/774 (2022.01); G06V 10/778 (2022.01)
CPC G06V 10/774 (2022.01) [G06V 10/778 (2022.01)] 15 Claims
OG exemplary drawing
 
1. A method for improving annotated datasets for training machine-learning algorithms, the method comprising:
i) annotating a dataset by a first machine-learning algorithm, wherein the dataset comprises a plurality of images and wherein the first machine-learning algorithm is configured to annotate the plurality of images by generating annotations for one or more features comprised in the plurality of images;
ii) training a second machine learning algorithm based on at least a first subset of the annotated dataset;
iii) evaluating the annotated dataset by:
a. using at least a second subset of the annotated dataset as an input dataset for the trained second machine learning algorithm in order to generate an output dataset comprising predictions of the one or more features; and
b. comparing the predictions generated by the trained second machine learning algorithm with the annotations generated by the first machine learning algorithm in order to extract an erroneous dataset from the input dataset, wherein the erroneous dataset comprises images associated with indications of annotation errors and/or missing annotations in the annotated dataset;
iv) re-annotating the erroneous dataset so to form a re-annotated dataset; and
v) repeating step ii) on the re-annotated dataset and repeating steps iii)-iv) until a number of images associated with indications of annotation errors and/or missing annotations is below a threshold.