International Display Workshops General Incorporated Association

10:20 〜 10:40

[FMC4-1 (Invited)] Display Defect Data Augmentation for Deep Learning Models

Changheon Lee1、JoonKyu Kim1、*Suk-Ju Kang1 (1.Sogang University (Korea))

Display Defect, Deep Learning, Data Augmentation

https://doi.org/10.36463/idw.2022.0275

This paper presents a deep learning-based data augmentation method for generating defect data. The generated data are to be used for training an anomaly detector for the purpose of detecting display defects. By comparing the generated data with those generated from previous methods we find that the deep learning-based data augmentation outperforms previous methods by producing photorealistic data covering a diverse range of real-world defects.