AI/Machine Learning Grading Prototype Almost Done

[kevinlenane]

Hey all - just to update those interested - we are getting closer to launching a proof of concept for a machine learning authentication, provenance(tampering) and grading app. I have the latter two working extremely well on 1982 Topps cards and the former working also super-well on 33 Goudeys (although admittedly I haven't come across any great reprint/fakes yet)

Anyway - I'm definitely going to demo it out here and let some folks play w it. The fingerprinting aspect has turned out to be really interesting in terms of a conditional and transactional history of a single card instance. I'm actually looking for any former graders or authenticators for some WFH contract work while we train our system. Open to hiring well paid short and long term arrangements - so if you have a background in grading please feel free to reach out. Otherwise stay tuned for automated card evaluation....

[kevinlenane]

Fwiw I also have it working on two modern sets 2018 Topps Living & Bowman Chrome. Once the algorithms start to get better w some real volume we won't really need to add sets individually but for the proof of concept stage we had to pick a few sets for near term demo purposes....

[G1911]

I'm on the side that generally sees grading as useless (beyond the $$$$) and easily corrupted by the inherent subjectivity allowed, but the realization of this concept is when it becomes interesting. Very interested in how this turns out and just what AI could and could not be taught to detect.

[Arazi4442]

Sounds exciting, please keep us updated.

When I got back into the hobby a few years back, I think one of the most surprising things was how much money is based solely on grading and all that grading was done exclusively by human opinions.

[brewing]

Sounds interesting. I'm looking forward to it.

[drcy]

The human element will never be and should never be removed, as grade is in part aesthetic and a matter of taste, and the human eye is a sophisticated scientific instrument. Also, what rules and weights (is centereing or color or edges most important, etc., 85/15 versus 80/20 for a grade, etc) are used to determine grade is subjective and that will never be unanimously or objectively answered.

However, automated and AI stuff would be excellent for consistency and for enforcing the rules.

I also know that AI will eventually be able to identify individual cards, such as a card that has been resubmitted.

[the 'stache]

Originally Posted by drcy

The human element will never be and should never be removed, as grade is in part aesthetic and a matter of taste, and the human eye is a sophisticated scientific instrument. Also, what rules and weights (is centereing or color or edges most important, etc., 85/15 versus 80/20 for a grade, etc) are used to determine grade is subjective and that will never be unanimously or objectively answered.

However, automated and AI stuff would be excellent for consistency and for enforcing the rules.

I also know that AI will eventually be able to identify individual cards, such as a card that has been resubmitted.

Pretty much exactly how I feel.

My greatest hope for this kind of technology would be that it nearly eliminates fraud from the hobby, as it applies to card alterations. If a high quality scan of a card could be performed, and the software could detect trimming (microscopic, irregular changes in edge consistency and quality), chemical alteration/cleaning, detecting paper stock inconsistencies (cards being rebuilt), incorrect ink being used-these last two to weed out reproductions being passed off as originals, etc-I'd be thrilled.

[swarmee]

Originally Posted by kevinlenane

Fwiw I also have it working on two modern sets 2018 Topps Living & Bowman Chrome.

Living set will be a very interesting challenge since it's only printed three cards at a time, may be cut or printed differently in each series of 3, and has been counterfeited. Will probably be more of a difficult set than you expected.