Back in the 90s I worked in the development dept of Epson's Scanner division. Fun times in those dot com days
Here's the thing gang -
Keeping it high level as my lights are flickering now and hoping I wont lose power as so many not far from me have.
Technology does exist that can be re-purposed for usage in detecting trimming. Actually extensively custom adapted or re-architectured would be more precise.
The same development that allows for Optical 3D Measurements are probably the key.
Advances in non-contact surface metrology, particularly those used for micron-level dimensional analysis, are likely the key enablers here.
A viable solution for trading cards would need to be developed from the ground up, albeit leveraging existing optical principles and hardware.
Thing is the amount of Development would be staggering. Right down to requiring many samples of all card issues then trimming them for comparative analysis. Years of effort and many cards would need to be harmed.
My wild A guess is 7 to low 8 figure range to get past a prototype phase
While it is plausible in 2026 to do what 1982FBWaxMemories has mentioned it's not currently realistic that PSA or anyone will spend the money. Re-couping those costs might be impossible.
Now in 5 years and factor in AI, instead of having to train humans, to interpret the data, and who knows!
Someone here also mentioned—though I can’t recall who—a database of high-resolution, high-DPI scans of card fronts and backs. That approach has actual merit. Such a database could be used to compare resubmitted cards against prior submissions to detect post-submission alterations. Host it on one of the many AWS database offerings (yes, I work there—somebody’s got to), layer in AI-based comparison models, and you’ve got a potentially scalable solution. There are most certainly many other methods worth exploring as well.
I’m no saint, and I’ve managed to totally offend many people here myself, but the level of acrimony in this thread—veiled or otherwise—is honestly striking. I assumed most of us here woudl be into a constructive discussion on identifying trimming and exchanging ideas
Idea Men matter just as much as those with technical expertise. In fact, at least for now, idea people may be harder to replace with AI.
@Groo said:
Back in the 90s I worked in the development dept of Epson's Scanner division. Fun times in those dot com days
Here's the thing gang -
Keeping it high level as my lights are flickering now and hoping I wont lose power as so many not far from me have.
Technology does exist that can be re-purposed for usage in detecting trimming. Actually extensively custom adapted or re-architectured would be more precise.
The same development that allows for Optical 3D Measurements are probably the key.
Advances in non-contact surface metrology, particularly those used for micron-level dimensional analysis, are likely the key enablers here.
A viable solution for trading cards would need to be developed from the ground up, albeit leveraging existing optical principles and hardware.
Thing is the amount of Development would be staggering. Right down to requiring many samples of all card issues then trimming them for comparative analysis. Years of effort and many cards would need to be harmed.
My wild A guess is 7 to low 8 figure range to get past a prototype phase
While it is plausible in 2026 to do what 1982FBWaxMemories has mentioned it's not currently realistic that PSA or anyone will spend the money. Re-couping those costs might be impossible.
Now in 5 years and factor in AI, instead of having to train humans, to interpret the data, and who knows!
Someone here also mentioned—though I can’t recall who—a database of high-resolution, high-DPI scans of card fronts and backs. That approach has actual merit. Such a database could be used to compare resubmitted cards against prior submissions to detect post-submission alterations. Host it on one of the many AWS database offerings (yes, I work there—somebody’s got to), layer in AI-based comparison models, and you’ve got a potentially scalable solution. There are most certainly many other methods worth exploring as well.
I’m no saint, and I’ve managed to totally offend many people here myself, but the level of acrimony in this thread—veiled or otherwise—is honestly striking. I assumed most of us here woudl be into a constructive discussion on identifying trimming and exchanging ideas
Idea Men matter just as much as those with technical expertise. In fact, at least for now, idea people may be harder to replace with AI.
Carry on folks and happy 2026 to all here!
Precise optical measurement hardly matters when the spec is random for cards cut by wire or blade. What are you comparing to.
As far as AI I really dislike using such a vacuous and overboard term to discuss computer vision technology. For grading a multiclass ML based on a vision transformer arch could be trained to perform qualitative analysis of sports cards given proper axial and coaxial lighting and frequency. Perhaps a binary classification for 10s and a multiclass with ordered classes. With normalized training data it’s possible you could build a passable network with a few million labeled images.
Using computer vision. Templating such as SIFT/SURF and matching to automate discovery but this only can detect the ones which have gone through grading unadulterated prior to being trimmed and resubmitted.
This is hardly constructive though as there are probably 100 people in this world who know how to build that model and none of them work for PSA.
Further PSA doesn’t have the images necessary to label the data. An on-axis image from a RGB camera isn’t gonna get this done.
That gives the most detail on a reasonable approach and I wouldn’t touch this without a 250 million budget. Good luck to PSA and anyone who thinks this is easy because…. AI. I think the Matrix is closer to reality.
Comments
Back in the 90s I worked in the development dept of Epson's Scanner division. Fun times in those dot com days
Here's the thing gang -
Keeping it high level as my lights are flickering now and hoping I wont lose power as so many not far from me have.
Technology does exist that can be re-purposed for usage in detecting trimming. Actually extensively custom adapted or re-architectured would be more precise.
The same development that allows for Optical 3D Measurements are probably the key.
Advances in non-contact surface metrology, particularly those used for micron-level dimensional analysis, are likely the key enablers here.
An example, and no this one wont work for Trading Cards of course, and silly to think it would. Its simply an example of an OK enough modern Optical 3D measurement device; https://www.jenoptik.us/products/optical-test-measurement-solutions/optical-3d-metrology
A viable solution for trading cards would need to be developed from the ground up, albeit leveraging existing optical principles and hardware.
Thing is the amount of Development would be staggering. Right down to requiring many samples of all card issues then trimming them for comparative analysis. Years of effort and many cards would need to be harmed.
My wild A guess is 7 to low 8 figure range to get past a prototype phase
While it is plausible in 2026 to do what 1982FBWaxMemories has mentioned it's not currently realistic that PSA or anyone will spend the money. Re-couping those costs might be impossible.
Now in 5 years and factor in AI, instead of having to train humans, to interpret the data, and who knows!
Someone here also mentioned—though I can’t recall who—a database of high-resolution, high-DPI scans of card fronts and backs. That approach has actual merit. Such a database could be used to compare resubmitted cards against prior submissions to detect post-submission alterations. Host it on one of the many AWS database offerings (yes, I work there—somebody’s got to), layer in AI-based comparison models, and you’ve got a potentially scalable solution. There are most certainly many other methods worth exploring as well.
I’m no saint, and I’ve managed to totally offend many people here myself, but the level of acrimony in this thread—veiled or otherwise—is honestly striking. I assumed most of us here woudl be into a constructive discussion on identifying trimming and exchanging ideas
Idea Men matter just as much as those with technical expertise. In fact, at least for now, idea people may be harder to replace with AI.
Carry on folks and happy 2026 to all here!
Precise optical measurement hardly matters when the spec is random for cards cut by wire or blade. What are you comparing to.
As far as AI I really dislike using such a vacuous and overboard term to discuss computer vision technology. For grading a multiclass ML based on a vision transformer arch could be trained to perform qualitative analysis of sports cards given proper axial and coaxial lighting and frequency. Perhaps a binary classification for 10s and a multiclass with ordered classes. With normalized training data it’s possible you could build a passable network with a few million labeled images.
Using computer vision. Templating such as SIFT/SURF and matching to automate discovery but this only can detect the ones which have gone through grading unadulterated prior to being trimmed and resubmitted.
This is hardly constructive though as there are probably 100 people in this world who know how to build that model and none of them work for PSA.
Further PSA doesn’t have the images necessary to label the data. An on-axis image from a RGB camera isn’t gonna get this done.
That gives the most detail on a reasonable approach and I wouldn’t touch this without a 250 million budget. Good luck to PSA and anyone who thinks this is easy because…. AI. I think the Matrix is closer to reality.
My favorite thing about this place is the pearl clutching and false outrage at banal remarks. Get. A. Grip.
Finally something we agree on, yes you do need to get grip on your manufactured outrage.
It's the singer not the song - Peter Townshend (1972)
Not even a minute do I buy the whole buh buh buh I'm a man-child japery - Me (2025)
You win. But I hope you learned something too.
no, no, you win
It's the singer not the song - Peter Townshend (1972)
Not even a minute do I buy the whole buh buh buh I'm a man-child japery - Me (2025)