Assigned MURA Defect Generation Based on Diffusion Model

Weizhi Liu, Chang Liu, Qiang Liu, Dahai Yu; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 4395-4402

Abstract


In this paper, we propose a novel method for assigned MURA generation using a diffusion model. MURA is a well-known problem in the display industry, which is difficult to be inspected because it is characterized by low contrast, blurry contours, blocky uneven brightness, and irregular shape patterns, and most defects have no rules to follow. Especially, for data-driven deep learning, the shortage of MURA samples collecting from the pipeline of manufactory is the first challenging problem, because the MURA sample happens with a low probability and in various ways. To relieve the problem, our proposed approach employs a diffusion model that generates MURA defect images using a few samples, which allows us to assign the position and class of MURA in the image. Specifically, our method leverages the diffusion process to estimate the visibility of MURA, which is then used to enhance the flexibility of the MURA detection process. We evaluate the performance of our method through MURA inspection. The results demonstrate the effectiveness of our proposed approach in addressing the MURA detection problem.

Related Material


[pdf]
[bibtex]
@InProceedings{Liu_2023_CVPR, author = {Liu, Weizhi and Liu, Chang and Liu, Qiang and Yu, Dahai}, title = {Assigned MURA Defect Generation Based on Diffusion Model}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2023}, pages = {4395-4402} }