guyiyifeurach

images example:  And red points are anomalies points.

In fact, I used object detection to solve the problem, but anomalies are different and can't be predicted,so detecting are not suitable way.In fact, I still insist that unsupervised learning...

There are different object categories in training data.  Such as SS(stainless steel)/glass/Al, all these small objects will show as 3~5mm black points with different textures.  But I need not classify...