From a field worker's perspective

I think the important point in "image processing inspection" is whether or not it can be thought of from the "viewpoint of the on-site worker".
In that light, isn't "simplicity" more important than anything else? If it is not easy, it will be impossible to understand what is happening, to make adjustments, to explain the results, and to trust the results. It may be true that using a difficult algorithm may increase detection capability. But if it is not easy to use, it is meaningless.
And "items judged to be good must be good." If only 9 out of 10 items can be inspected, there is no guarantee that they are good, even if they are judged to be good in that inspection. The remaining one item, i.e., the item that is difficult to inspect by image processing, is also difficult to inspect by visual inspection, and in the end, it is a common story that no labor saving was achieved. In the end, we believe that "all items inspection" is a must.
Furthermore, "the price must match the return on investment." Assuming that the system works 100% as expected, I think that the limit for the image processing inspection part is 2 million yen. In this light, I think it is true that the pricing of "image processing equipment" sold as a finished product is exquisite, while "customized image processing equipment" is quite expensive. However, when I think about whether it "works as expected" or not....
Anyway, "on-site". With that in mind, we are developing/improving FlexInspector.