Quote:
Originally Posted by Musashi
Neither of the two biggest challenges for AI grading are the coding (which isn't up to par yet, but will likely get there sooner rather than later). Rather, they are
1) Insufficient examples in each grade (especially for scarcer issues) to properly train the model. That's why none of the smaller companies trying to give digital grading a go (that I'm aware of) are doing vintage.
2) Will also require a leap forward in scanning technology. Surface wrinkles and light creases tend to disappear in scans made even by quality scanners. System can't grade what it can't see.
|
1) I think if someone wanted to automate it would require splitting the grade, something I've always favored.
Automating how well the card is preserved I think is possible right now. There are scanning devices that generate a monochrome image of something in detail. From that you can get size, corner wear, creases, and surface. You don't really even need to know what set the card is from, other than to compare size. Picking up trims should also be possible for most sets.
Other things that influence the grade like centering and print flaws and stains etc are a different matter. My wife is in software and assures me that centering would be very easy given scans that were consistent. She also thinks that identification and spotting other flaws that would affect the grade are fairly easy. I have doubts...
2) A high dpi scan from a good scanner reveals a LOT of things, even stuff you can't readily see. 1200 dpi is amazing. But it's also slow and takes a lot of memory/storage