Asking for advices

[mathispezet]

Hello ! Firstable congratulations for your great work ! I'm trying to do anomaly detection on industrial metallurgic micrographics images and i think Glass is the solution i need, it's quite tough ( at least for me ) and i was wondering if you maybe had some advices on how i should use GLASS for this. I'm new to do data science and i'm a bit struggling with all of this.

[cqylunlun]

Thank you for your recognition. The main issue with training on a custom dataset lies in the dataset format. Please refer to issues #50 and #48, and configure it according to your specific scenario.

[mathispezet]

Thank you for your answer ! I managed to run training on a custom dataset, but I was mostly wondering if this solution is really suitable for my problem. I need to detect defects on texture contour images, rather than on a global image of the part. Since the part pose is not normalized, the images are never the same from one part to another. Does GLASS handle spatial texture shifts in the images? Does this negatively affect training quality?

[cqylunlun]

Based on the datasets we have tested, GLASS can handle periodic texture patterns and moderate spatial shifts in defect detection. However, the exact performance in your scenario will still depend on your experimental results.