WO2025094053 - DATASET-LEVEL SOCIETAL BIAS MITIGATION WITH TEXT-TO-IMAGE MODEL
National phase entry:
Publication Number
WO/2025/094053
Publication Date
08.05.2025
International Application No.
PCT/IB2024/060656
International Filing Date
29.10.2024
Title **
[English]
DATASET-LEVEL SOCIETAL BIAS MITIGATION WITH TEXT-TO-IMAGE MODEL
[French]
ATTÉNUATION DE BIAIS SOCIÉTAL AU NIVEAU D'UN ENSEMBLE DE DONNÉES AVEC MODÈLE TEXTE VERS IMAGE
Applicants **
SONY GROUP CORPORATION
Inventors
HIROTA, Yusuke
ANDREWS, Jerone
ZHAO, Dora
PAPAKYRIAKOPOULOS, Orestis
MODAS, Apostolos
XIANG, Alice
Priority Data
63/595,656
02.11.2023
US
18/886,531
16.09.2024
US
Application details
| Total Number of Claims/PCT | * |
| Number of Independent Claims | * |
| Number of Priorities | * |
| Number of Multi-Dependent Claims | * |
| Number of Drawings | * |
| Pages for Publication | * |
| Number of Pages with Drawings | * |
| Pages of Specification | * |
| * | |
| Number of Office Actions | * |
| * | |
International Searching Authority |
EPO
* |
| Recordal of a Change of the Applicant's Name/Address |
Change of Applicant's Name and Address
* |
| Type of Assignment |
The Standard Agent's Assignment
* |
| Applicant's Legal Status |
Legal Entity
* |
| * | |
| * | |
| * | |
| * | |
| * | |
| Entry into National Phase under |
Chapter I
* |
| Patent Delivery |
Send the Letters Patent by Courier
* |
| Translation |
|
* The data is based on automatic recognition. Please verify and amend if necessary.
** IP-Coster compiles data from publicly available sources. If this data includes your personal information, you can contact us to request its removal.
Quotation for National Phase entry
| Country | Stages | Total | |
|---|---|---|---|
| China | Filing, Examination, Granting | 2275 | |
| EPO | Filing, Examination, Granting | 10853 | |
| Japan | Filing, Examination, Granting | 2337 | |
| South Korea | Filing, Examination, Granting | 2382 | |
| USA | Filing, Examination, Granting | 4740 |
Total:
22,587
The term for entry into the National Phase has expired. This quotation is for informational purposes only
Contact Us
Abstract[English]
Systems and methods are used to mitigate societal bias in image-text datasets by removing spurious correlations between protected groups and image attributes. Using text-guided inpainting models, the methods ensures protected group independence from all attributes and mitigates inpainting biases through data filtering. Evaluations on multi-label image classification and image captioning tasks show that the methods effectively reduce bias without compromising performance across various models.[French]
La présente invention porte sur des systèmes et des procédés qui sont utilisés pour atténuer un biais sociétal dans des ensembles de données d'image-texte par élimination de corrélations parasites entre des groupes protégés et des attributs d'image. À l'aide de modèles de retouche d'image guidés par texte, les procédés assurent une indépendance de groupe protégée à partir de tous les attributs et atténuent les biais de retouche d'image par filtrage de données. Des évaluations sur une classification d'image à étiquettes multiples et des tâches de réalisation de légende d'image montrent que les procédés réduisent efficacement le biais sans compromettre les performances de divers modèles.