A discussion about using AI to grade comic books. 3 3

[Buzzetta]

On 12/24/2023 at 11:09 AM, Buzzetta said:

Remember that when it comes to third party grading that baseball cards have always been on the forefront of the technology, presumably due to lack of complexity in grading a card compared to other collectables.  AI is making an appearance in card grading.   As the technology progresses, it is only a matter of time before some aspects will find their way into grading other collectables. 

https://agscard.com/

Now whether or not it is successful remains to be seen.  However the tech has been implemented. 

 

On 12/24/2023 at 11:14 AM, VintageComics said:

It's been implemented in cards for 2 specific reasons that I can think of.

1) cards are much easier to grade than books

2) the money in cards dwarfs the money in comic books. Remember when $1 million comic book was crazy money? There were already multi-million $ cards floating around. 

Again, I'm not saying AI will never grade comics, I'm just saying making it accurate, cost efficient and time efficient are feasible over all, are the complicating factors. 

I already said that in the quote that you are responding to and "correcting" me on. 

The technology will develop.  It will become cheaper.    Remember, at one point no one envisioned many other fine tune automated tech from bottling to applying the retinas and other paint apps on 3.75" toys on a mass market scale.   It's coming. 

[VintageComics]

On 12/24/2023 at 11:44 AM, Buzzetta said:

 

I already said that in the quote that you are responding to and "correcting" me on. 

The technology will develop.  It will become cheaper.    Remember, at one point no one envisioned many other fine tune automated tech from bottling to applying the retinas and other paint apps on 3.75" toys on a mass market scale.   It's coming. 

I didn't correct you, I added to your points.

Again, nobody stated it isn't coming. I'm simply saying it isn't coming soon and the reason I believe this, is for the reason that the intersections of efficiency (time) / complexity / cost / profitability don't make it feasible yet.

Painting a 3.75" toy for child consumption is very different than flipping through a vintage book, imaging 30+ pages, programming software to identify every possible option and formulating an accurate grade. 

Like someone stated above, just a speck of dust can change everything, whereas humans would factor that in quickly. 

There is a sliding scale between reliability and accuracy and the human relationships on this and computer relationships on this operate very differently. I think the average Joe thinks they're the same. They're not. 

^

That was me correcting you. 

-------------------------------------------

As I stated in the #252 thread, I believe this will become reality in the future, around the time when more complex bio-digital convergence in humans. No, not hearing aids - I'm talking about intelligent bio-digital convergence.

Edited December 24, 2023 by VintageComics

[Buzzetta]

On 12/24/2023 at 12:03 PM, VintageComics said:

I didn't correct you, I added to your points.

Again, nobody stated it isn't coming. I'm simply saying it isn't coming soon and the reason I believe this, is for the reason that the intersections of efficiency (time) / complexity / cost / profitability don't make it feasible yet.

Painting a 3.75" toy for child consumption is very different than flipping through a vintage book, imaging 30+ pages, programming software to identify every possible option and formulating an accurate grade. 

I was commenting on the advancements of technology.   Try and follow along.  So, since 2020-2021 they started introducing something labeled "photo real" technology on 3.75" and 6" figures.  For years before that they were unable to align things like eyeballs and retinas as well as the shading on the faces of figures.   The whole process involves scanning individuals when possible or creating a 3D composite based on a 2D images lifted from media.   The tech is getting frighteningly advanced from where we were three years ago.   The optics and application are set in place as seen with the toy industry.   Now, if they are able to do that for a $16.99 (3.75" going rate) and $24.99 (6" going rate) action figure imagine what they could develop for industrial consumption that is of repeated use. 

 

On 12/24/2023 at 12:03 PM, VintageComics said:

Like someone stated above, just a speck of dust can change everything, whereas humans would factor that in quickly. 

There is a sliding scale between reliability and accuracy and the human relationships on this and computer relationships on this operate very differently. I think the average Joe thinks they're the same. They're not. 

^

That was me correcting you. 

-------------------------------------------

As I stated in the #252 thread, I believe this will become reality in the future, around the time when more complex bio-digital convergence in humans. No, not hearing aids - I'm talking about intelligent bio-digital convergence.

We are starting to see it in grading cards.  So as I suggested, the first steps will be in developing AI in assessing page quality in so far as, "What is white compared to OW to White?"  That is rather simplistic and would create far greater reliability than what the graders are able to accomplish in this area.  If they adopt some of the optics tech and fine tune it I believe that the tech is already there to discern nuances in color across the page.   It is a matter of putting it all together. 

[universal soldier]

I can see this happening at some point (likely sooner than later) but two things stick out to me that would need to be answered before this would be efficient and effective. 

The scans need to be done post arrival as it would be too easy to manipulate and/or fake coming from the submitter. An actual human would need to unpack, scan and pre-process etc... so how would this save time and effort? With all the QC issues you wouldn't want anyone but the grader or a trained employee performing this type of task.

I've seen a few posts taking about three dimensional views. Certainly something like this is needed as even a high res scan may not show a slight NCBC or two that can only be seen at an angle or a subtle waviness that can be the difference bewteen a 9.6 or 9.8 etc.. How close are we to this type of tech?

Edited December 24, 2023 by universal soldier
speling ishard

[VintageComics]

On 12/24/2023 at 12:52 PM, Buzzetta said:

I was commenting on the advancements of technology.   Try and follow along.  So, since 2020-2021 they started introducing something labeled "photo real" technology on 3.75" and 6" figures.  For years before that they were unable to align things like eyeballs and retinas as well as the shading on the faces of figures.   The whole process involves scanning individuals when possible or creating a 3D composite based on a 2D images lifted from media.   The tech is getting frighteningly advanced from where we were three years ago.   The optics and application are set in place as seen with the toy industry.   Now, if they are able to do that for a $16.99 (3.75" going rate) and $24.99 (6" going rate) action figure imagine what they could develop for industrial consumption that is of repeated use. 

You're painting inanimate objects that aren't easily damaged first, and the objects are 3 dimensional, but it's basically a 2 dimensional canvas.

We're talking about delicate comics that bend an rip - they don't stay straight or motionless and you'd need to image each page from all angles (or do it simultaneously - whatever, different discussion).

Assembly lines have been painting cars for decades. 

Robots have been used to stitch upholstery forever. 

The comic grading equivalent here would be to paint the inside and outside of a complex object with fine linework, precisely every time. Or something like that. 

The LAYERS of complexity is what in my POV, makes grading a comic different than just a two dimensional problem to solve on an inanimate object that can't be damaged easily.

You have to undo ALL of those layers and solve them. 

On 12/24/2023 at 12:52 PM, Buzzetta said:

We are starting to see it in grading cards.  So as I suggested, the first steps will be in developing AI in assessing page quality in so far as, "What is white compared to OW to White?"  That is rather simplistic and would create far greater reliability than what the graders are able to accomplish in this area.  If they adopt some of the optics tech and fine tune it I believe that the tech is already there to discern nuances in color across the page.   It is a matter of putting it all together. 

We're also starting to see it in comics. CGC has already admitted to implementing AI in some of it's procedures. 

Nobody is arguing against consistency. I've already stated multiple times that certain aspects of grading, like estimating PQ or assigning a pre-grade range is very likely and sooner. 

What is being argued is FEASIBILITY. 

In your own words:

On 12/24/2023 at 12:52 PM, Buzzetta said:

Try and follow along.

 

[VintageComics]

On 12/24/2023 at 1:17 PM, universal soldier said:

I've seen a few posts taking about three dimensional views. Certainly sometihng like this is needed as even a high res scan may not show a slight NCBC or two that can only be seen at an angle or a subtle waviness that can be the differenct bewteen a 9.6 or 9.8 etc.. How close are we to this type of tech?

This sounds like a great place to ease the conversation into.

[shadroch]

On 12/24/2023 at 9:14 AM, VintageComics said:

It's been implemented in cards for 2 specific reasons that I can think of.

1) cards are much easier to grade than books

2) the money in cards dwarfs the money in comic books. Remember when $1 million comic book was crazy money? There were already multi-million $ cards floating around. 

Again, I'm not saying AI will never grade comics, I'm just saying making it accurate, cost efficient and time efficient are feasible over all, are the complicating factors. 

Today's system is not accurate,  nor cost or time efficient.

[VintageComics]

On 12/24/2023 at 2:33 PM, shadroch said:

Today's system is not accurate,  nor cost or time efficient.

That statement is misleading at best and not true at all, at worst. 

It's more accurate than the average person which is what it's supposed to be. 

It's not perfect, but then neither is NASA. 

If they had a more accurate, cost and time efficient method do you think they'd be using it?

It's a question with a yes or no answer. 

[Stefan_W]

If the technology is available at a price point that is not prohibitive I am sure AI will be used.

In the mean time, I think discussions about this topic will boil down to whether or not people want to see AI used in grading comics. I am on the fence because while the consistency that may be offered by AI is really attractive, particularly with modern books that have very few flaws, I am skeptical with using this for all comics because for some (say mid grade silver age books) I would rather have someone independently assess it than try to train a program to take hundreds or more factors associated that book into account. 

I am open enough either way to have my mind changed as technology develops and rational enough to understand that my opinion will not matter in the slightest when final decisions on this matter are made. 

[FlyingDonut]

On 12/24/2023 at 11:06 AM, VintageComics said:

Just the topic of scanning and interpreting images, (and this post doesn't even touch on how the book will be manipulated for scanning) is pretty complex. 

The problem as I perceive it it isn't that AI can't just scan and read images, we all know it can, but a book is a 3 dimensional item. 

It's a combination of things like how will the book be delivered and show to the AI program / scanner?

Who will expose both sides of every page in a book (16 - 32+ wraps or 32 to 64+ pages?)

How far open do the pages need to be?

Can you scan them accurately at an angle?

Can you scan them accurately with curvature on the pages (meaning they aren't laying flat)?

How do you inspect staples? Especially in square bounds?

You're already imaging 32-64 pages, how do you image the sides of the book, like the open edge, the spine the tops and bottoms?

I've had books with defects ONLY apparent on the open edge of the book - literally across the face of the page ends. Really. I had a book where someone wrote in ink that only showed up in UV on the literal edges of the pages and the defect wasn't visible from the flat side of the page. 

This was one of those strange defects that I was talking about that you would have to completely retrain AI for once you uncovered a new flaw. 

Then we have the problems of stuff like specks of dust and other things that can corrupt images as discussed above. 

Every little new factoid is a learning curve for AI that would take countless hours to adapt to, that humans can factor in, in a matter of seconds through a personal discussion around a grading table. 

 

These are just some of the things that I've written "off the cuff" so to speak. There are probably a zillion others. 

 

I work for a firm that is pretty far out on the edge of AI so I'm pretty aware of what is out there because I'm trying to sell it.

Everything in this post can be easily fixed with data and changes. AI learns. That's the whole point. There is already AI doing things that were "impossible" 18 months ago and the curve is moving extremely rapidly.

If you think AI couldn't be adopted to grade comics, you're just flat out wrong. The reason why it wouldn't happen has nothing to do with the software/how it would work, it wouldn't happen because there's currently no economic reason to do it.

[Buzzetta]

On 12/24/2023 at 2:25 PM, VintageComics said:

You're painting inanimate objects that aren't easily damaged first, and the objects are 3 dimensional, but it's basically a 2 dimensional canvas.

We're talking about delicate comics that bend an rip - they don't stay straight or motionless and you'd need to image each page from all angles (or do it simultaneously - whatever, different discussion).

Assembly lines have been painting cars for decades. 

Robots have been used to stitch upholstery forever. 

The comic grading equivalent here would be to paint the inside and outside of a complex object with fine linework, precisely every time. Or something like that. 

The LAYERS of complexity is what in my POV, makes grading a comic different than just a two dimensional problem to solve on an inanimate object that can't be damaged easily.

You have to undo ALL of those layers and solve them. 

We're also starting to see it in comics. CGC has already admitted to implementing AI in some of it's procedures. 

Nobody is arguing against consistency. I've already stated multiple times that certain aspects of grading, like estimating PQ or assigning a pre-grade range is very likely and sooner. 

What is being argued is FEASIBILITY. 

In your own words:

 

The tech is getting there.    You sound like Blockbuster who said that was Netflix wanted to do was not feasible. 

One day there will be AI to handle some of the grading chores.  

Now whether or not you see that happening, kind of does not matter. 

[THE_BEYONDER]

On 12/24/2023 at 12:03 PM, VintageComics said:

I didn't correct you, I added to your points.

Again, nobody stated it isn't coming. I'm simply saying it isn't coming soon and the reason I believe this, is for the reason that the intersections of efficiency (time) / complexity / cost / profitability don't make it feasible yet.

Painting a 3.75" toy for child consumption is very different than flipping through a vintage book, imaging 30+ pages, programming software to identify every possible option and formulating an accurate grade. 

Like someone stated above, just a speck of dust can change everything, whereas humans would factor that in quickly. 

There is a sliding scale between reliability and accuracy and the human relationships on this and computer relationships on this operate very differently. I think the average Joe thinks they're the same. They're not. 

^

That was me correcting you. 

-------------------------------------------

As I stated in the #252 thread, I believe this will become reality in the future, around the time when more complex bio-digital convergence in humans. No, not hearing aids - I'm talking about intelligent bio-digital convergence.

Is AI grading comics considered science fiction?

[shadroch]

On 12/24/2023 at 12:41 PM, VintageComics said:

That statement is misleading at best and not true at all, at worst. 

It's more accurate than the average person which is what it's supposed to be. 

It's not perfect, but then neither is NASA. 

If they had a more accurate, cost and time efficient method do you think they'd be using it?

It's a question with a yes or no answer. 

 

I believe they are more concerned with making a profit than being accurate, or efficient.  Not that there is anything wrong with that. 

Edited December 24, 2023 by shadroch

[namisgr]

On 12/24/2023 at 1:17 PM, universal soldier said:

I've seen a few posts taking about three dimensional views. Certainly something like this is needed as even a high res scan may not show a slight NCBC or two that can only be seen at an angle or a subtle waviness that can be the difference bewteen a 9.6 or 9.8 etc.. How close are we to this type of tech?

It's been in use for a long time now, for instance in medical imaging and biological research involving microscopy.

After head on images are captured of the front and back covers along with the pages, angled images could be easily obtained, adjusting the angle between the image sensor and the book either by tilting the sensor or the book.

Edited December 24, 2023 by namisgr

[Prince Namor]

On 12/25/2023 at 3:02 AM, FlyingDonut said:

I work for a firm that is pretty far out on the edge of AI so I'm pretty aware of what is out there because I'm trying to sell it.

Everything in this post can be easily fixed with data and changes. AI learns. That's the whole point. There is already AI doing things that were "impossible" 18 months ago and the curve is moving extremely rapidly.

If you think AI couldn't be adopted to grade comics, you're just flat out wrong. The reason why it wouldn't happen has nothing to do with the software/how it would work, it wouldn't happen because there's currently no economic reason to do it.

This sounds like a great place to ease the conversation into.

[VintageComics]

On 12/24/2023 at 3:02 PM, FlyingDonut said:

I work for a firm that is pretty far out on the edge of AI so I'm pretty aware of what is out there because I'm trying to sell it.

Everything in this post can be easily fixed with data and changes. AI learns. That's the whole point. There is already AI doing things that were "impossible" 18 months ago and the curve is moving extremely rapidly.

If you think AI couldn't be adopted to grade comics, you're just flat out wrong. The reason why it wouldn't happen has nothing to do with the software/how it would work, it wouldn't happen because there's currently no economic reason to do it.

Or it can't be done economically yet. 

NASA could probably build a system that works reasonably well, but it wouldn't be for public consumption at NASA levels.

Could it do it fast enough? Great topic for discussion. 

You seem very well versed.

So how do they hold the book for imaging, flip the pages and how long would each book take? And how accurate would it be?

[VintageComics]

On 12/24/2023 at 3:17 PM, Buzzetta said:

One day there will be AI to handle some of the grading chores.  

That's not what the discussion is about.

Almost everyone agrees it can or will be able to do SOME of the grading chores. 

This discussion is about whether it can do ALL of the grading chores and if it's feasible (cost efficient / profitable / accurate / quick enough).

[VintageComics]

On 12/24/2023 at 4:05 PM, namisgr said:

After head on images are captured of the front and back covers along with the pages,

Question 1) How do you capture head on images of pages unless you open the book 30-60 times at a 45 degree angle or greater?

Question 2) Who or what will do the opening?

On 12/24/2023 at 4:05 PM, namisgr said:

angled images could be easily obtained, adjusting the angle between the image sensor and the book either by tilting the sensor or the book.

Question 3) You'd need another 30-60 images of each page at an angle then?

Question 4) How much time do you foresee this taking just for one book?

[FlyingDonut]

On 12/24/2023 at 10:41 PM, VintageComics said:

Question 1) How do you capture head on images of pages unless you open the book 30-60 times at a 45 degree angle or greater?

Question 2) Who or what will do the opening?

Question 3) You'd need another 30-60 images of each page at an angle then?

Question 4) How much time do you foresee this taking just for one book?

These are not AI problems, they're hardware problems, but ones that could be initially handled through an human operator opening everything and imaging them. Imaging items in a high resolution and high-quality environment is not difficult. Google has been doing it for years and their systems are getting better and better.

None of the things you have talked about in this thread (or the other one) are AI issues. If someone wanted to find an econometric model that worked AND was willing to invest in fixing the hardware problem, AI grading of collectibles would be very quick to get to a very high level of confidence - its just learning and continual examples teaching the system how it works. Grading cards using AI is probably coming because the hardware problems would go away.

[CAHokie]

If we can map a brain we map some comic book pages. It would learn the more it does. I believe the image capturing would be able to grab everything it needs from normal speed turning of pages.