US 2026/0301387 A1Published application

Percentile-based pseudo-label selection for multi-label semi-supervised classification

AssigneeAutodesk, Inc.

InventorsJunxiang Huang, Alexander Huang, Beatriz Chinelato Guerra, Yen-Yun Yu

Life of this patent

Estimating the term…

Loading patent…

Abstract

A method and system provide for augmenting a photograph. An unlabeled photograph is obtained. A weakly augmented photograph and a strongly augmented photograph are obtained from the unlabeled photograph based on different types of data augmentation methods. The weakly augmented photograph is processed through a model to generate multiple weakly augmented photograph class predictions (with assigned probabilities). The multiple weakly augmented photograph class predictions are converted into positive pseudo-labels (indicating a presence of a class) or negative pseudo-labels (indicating absence of a class) using different fixed percentile thresholds. The strongly augmented photograph is processed through the model to generate a strongly augmented photograph class prediction. The model is trained to make the strongly augmented photograph label prediction match the positive pseudo-label via a cross-entropy loss. The trained model is then utilized to label the unlabeled photograph with multiple labels.

Claims20

Dependent claims

A computer-implemented method for augmenting a photograph, comprising:

(a) obtaining an unlabeled photograph;
(b) obtaining a weakly augmented photograph based on the unlabeled photograph, wherein the weakly augmented photograph is based on a first type of data augmentation;
(c) obtaining a strongly augmented photograph based on the unlabeled photograph, wherein:
(i) the strongly augmented photograph is based on a second type of data augmentation;
and
(ii) the strongly augmented photograph is more strongly augmented compared to that of the weakly augmented photograph;
(d) processing the weakly augmented photograph through a model to generate multiple weakly augmented photograph class predictions, wherein each weakly augmented photograph class prediction is assigned a probability by the model;
(e) converting the multiple weakly augmented photograph class predictions into positive pseudo-labels or negative pseudo-labels, wherein:
(i) a first class prediction, of the weakly augmented photograph class predictions, is converted to the positive pseudo-label when the probability for the first class prediction is above a positive score threshold, and wherein the positive pseudo-label indicates a presence of a class;
and
(ii) a second class prediction, of the weakly augmented photograph class predictions, is converted to the negative pseudo-label when the probability for the second class prediction is below a negative score threshold, and wherein the negative pseudo-label indicates an absence of a class;
(f) processing the strongly augmented photograph through the model to generate a strongly augmented photograph class prediction;
(g) training the model to make the strongly augmented photograph class prediction match the positive pseudo-label, wherein the training is based on the positive pseudo-labels and the negative pseudo-labels;
and
(h) utilizing the model to label the unlabeled photograph with multiple labels.

12 more elements

A computer-implemented system for augmenting a photograph, comprising:

(a) a computer having a memory;
(b) a processor executing on the computer;
(c) the memory storing a set of instructions, wherein the set of instructions, when executed by the processor cause the processor to perform operations comprising:
(i) obtaining an unlabeled photograph;
(ii) obtaining a weakly augmented photograph based on the unlabeled photograph, wherein the weakly augmented photograph is based on a first type of data augmentation;
(iii) obtaining a strongly augmented photograph based on the unlabeled photograph, wherein:
(1) the strongly augmented photograph is based on a second type of data augmentation;
and
(2) the strongly augmented photograph is more strongly augmented compared to that of the weakly augmented photograph;
(iv) processing the weakly augmented photograph through a model to generate multiple weakly augmented photograph class predictions, wherein each weakly augmented photograph class prediction is assigned a probability by the model;
(v) converting the multiple weakly augmented photograph class predictions into positive pseudo-labels or negative pseudo-labels, wherein:
(1) a first class prediction, of the weakly augmented photograph class predictions, is converted to the positive pseudo-label when the probability for the first class prediction is above a positive score threshold, and wherein the positive pseudo-label indicates a presence of a class;
and
(2) a second class prediction, of the weakly augmented photograph class predictions, is converted to the negative pseudo-label when the probability for the second class prediction is below a negative score threshold, and wherein the negative pseudo-label indicates an absence of a class;
(vi) processing the strongly augmented photograph through the model to generate a strongly augmented photograph class prediction;
(vii) training the model to make the strongly augmented photograph class prediction match the positive pseudo-label, wherein the training is based on the positive pseudo-labels and the negative pseudo-labels;
and
(viii) utilizing the model to label the unlabeled photograph with multiple labels.

15 more elements

Dependencies are inferred from the claim text.

Facts

Application: 19/695,945Filed 2 Jun 2026
Priority: 26 Aug 2022
Published: 1 Oct 2026
Term ends: Not estimatedThe source records don’t support an estimate.

Classification3 codes

G06V 10/7753
Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting: incorporation of unlabelled data, e.g. multiple instance…
G06V 10/764
Using pattern recognition or machine learning: using classification, e.g. of video objects
G06V
Image and video recognition

Family

Application63/373,588Us provisional application · filed 26 Aug 2022
Application18/452,780Continuation · filed 21 Aug 2023
Application19/695,945Continuation
US 12,657,887 B2Related patent
US 2024/0071056 A1Published application
US 2026/0301387 A1This patent

From the US filing history. Foreign family members aren’t listed.