Video of automated surface card grading in action

[kevinlenane]

Moving to existing thread

[MuncieNolePAZ]

Thanks for sharing.

[teza11]

Interesting. How repeatable is the machine score? For example, if you ran the analysis 5 times, does the same machine score always appear?

Jeff

[rdwyer]

Bye Bye PSA.

[Fuddjcal]

Originally Posted by kevinlenane

As promised - i thought id include a basic demo of surface grading which is basically taking clusters of color anomalies in the pixels, ruling out various clusters which are generally (at a set level) deemed okay by professional grader viewpoints (print marks etc.) and then docking accordingly based on what remains in terms of anomalies in the card surfaces. I'm adding in the easier sub-grades this week but I thought some of the folks here would get a kick out of seeing a machine grade the surface of the cards even though its in a basic command line right now...

https://www.youtube.com/watch?v=OYsBRuU1qNs&t=90s

That's the way!!!! Very nice. Hope it makes a splash and can take into account trimming and toluene baths someday?

[swarmee]

Originally Posted by teza11

Interesting. How repeatable is the machine score? For example, if you ran the analysis 5 times, does the same machine score always appear?

Especially with a different input image of the same card.

[Bigdaddy]

So is this looking mostly at colors or is it somehow looking a three dimensional attributes - say an indentation in the card?

Also, are the detractions things that would have been present on Day 1 - when it was pulled from a pack in 1982 or are they generally things that happened to the card during the ensuing 38 years?

BTW, I really like what you are doing here.

[swarmee]

Originally Posted by Bigdaddy

So is this looking mostly at colors or is it somehow looking a three dimensional attributes - say an indentation in the card.

This is why you always add the information to the long thread already about this... maybe Leon can merge them?

[hcv123]

Keep up the great work!

No such thing as a "perfect system", but this seems like it will be a heck of a lot more consistent than humans!

[oldeboo]

Out with the old, in with the new, the future is here. There's a new sheriff in town. Great video.

[insidethewrapper]

How will you know if this system is better than professional graders ? Based on the information you have shared so far it is about the same. Will you determine the system is valid if it relates closely to the professional graders or if there is a big difference etc . ? So you also need good reliability ( use same card and get same grade ) and validity . How are the results so far with reliability ? Good Luck on this venture. Corners and borders for most cards should be easier I would think to program into grades.

[GasHouseGang]

If machine grading ever gets to market I wonder if they will put the software version used to create the grade on the slab? Because software always can improve and engineers are never satisfied, I would imagine that the software will continue to get upgraded regularly unless they are forced to lock it down and have strict configuration control. This usually doesn't happen in private industry.

I can imagine collectors in the future arguing about which software version was used to grade the card they are trying to buy. Maybe it might look something like this:

[Bobbycee]

How does the computer figure in the all important eye appeal?

[rdwyer]

So far this is the first step. I'm sure that they'll include centering, creases, pinholes, scrapes, and everything else in the future. It'll probably be a 100 step process that'll cover every aspect of card grading before they're done. This is a great start so far. Looking forward to future updates.

[Bigdaddy]

Originally Posted by GasHouseGang

If machine grading ever gets to market I wonder if they will put the software version used to create the grade on the slab? Because software always can improve and engineers are never satisfied, I would imagine that the software will continue to get upgraded regularly unless they are forced to lock it down and have strict configuration control. This usually doesn't happen in private industry.

I can imagine collectors in the future arguing about which software version was used to grade the card they are trying to buy. Maybe it might look something like this:

That is important and most certainly the software will change with time, however the current TPGs have changed their grading standards over time with no noted information on the flip. Other than the comment 'That is an old PSA slab and that card would never get that grade if re-subbed today.'

[kevinlenane]

I definitely should have added to the existing thread - my bad! So with different images of the same card the software should pick up on the finger print and re assess condition while coming back with the same grade. I'll actually be able to love demo shorty with the rest of the grades so we I'll be able to show that test w but more thoroughly. It will take some time to get the bulk of the set support done but that's just a matter of a few months with training data. I will day surface is by far the hardest but is definitely an exercise in breaking any imperfection into color changes and then training it on what irregularities are acceptable to someone with professional grading experience. I'll plan on reaching out to a few folks with an actual demo at the beginning of next week to get some additional feedback.