-
When you click on links to various merchants on this site and make a purchase, this can result in this site earning a commission. Affiliate programs and affiliations include, but are not limited to, the eBay Partner Network.
-
Posts
132 -
Joined
-
Last visited
Content Type
Forums
CGC Journals
Gallery
Events
Store
Posts posted by RhialtoTheMarvellous
-
-
Grade a comic using machine learning
The tooling for building out Machine Learning models has become fairly easy to use and I wanted to see if I could build a model that can grade a comic. I've got the tooling to create the model, but what I need are images.As a start I wanted to focus on one particular comic book. The reason I only want to use one book is because of the evaluation criteria. The idea is that we want to be able to classify a book into a particular score. So the criteria is much more narrow than for instance if we were going to recognize whether a picture has a comic book in it. That case would be relatively easy to model.
I was thinking ASM #300 would probably be something people could dig up easily in various grades. Granted we won't probably see any 2.0s, but we can probably get a decent sample of images from 7.0 to 9.8.
To be clear, this is not a replacement for the CGC service and I have no intent to make any of this into a pay to play product. I honestly have no idea whether it will work or not due to the image sample requirements and it's difficult to create something like this just for one comic book much less the millions of them that exist out there. Given the variety of defects possible I will likely need many samples at each score level to get something that can discern accurately between a 9.4 and a 9.8.
What I need-
Raw front and back cover scans of ASM #300. One of each per copy. I must have both the front and back cover.
- Obviously this could be difficult to find since people don't always scan a comic front and back before getting it graded.
- The grade assigned to those front/back scans as a number (9.8, 9.7, 2.0, etc), no text grades please.
Image caveats
- No cropping of edges, the comic front and back cover must be fully visible.
- No partial images (ie corner shots, close ups of defects etc).
-
The defects that cause the grade issues must be visible in the scan.
- This won't work for interior page defects. Keeping it simple for now.
- It won't work for books already in a holder.
- Try to keep them reasonably sized.
- JPG or PNG only.
The more scans at each grade that I can obtain the better the model will end up being.
If it ends up that it has decent accuracy I'll make the model publicly available and put up a web page where you can submit your own images of ASM 300 for grading. If anything it might be a halfway decent pre-screen.
-
Raw front and back cover scans of ASM #300. One of each per copy. I must have both the front and back cover.
-
I wouldn't crack it. That area by the staples looks more like color rub rather than color transfer.
-
That's not NM+ (ie 9.6), it's not even close.
You can't remove pen from the cover (at least not without it being detected). It looks like there is color rub around the area where someone tried. That and the tear and chips at the top I'd say knocks it down into 8.0-8.5 territory.At $250 US if the ASM 252 is actually NM, you probably didn't get ripped, but it's not a deal.
-
-
8.0
Front cover bottom right corner lets it down.
Looks like there are some chips coming off the top of the front cover as well, but that could be the lighting or shadows.
The back cover on the bottom looks like it might have light moisture staining, but it could just be the lighting. If it is moisture then that knocks it down more.
-
If those were just dents and didn't have color breaking scrapes in them you could definitely remove them with various pressing techniques, but when the color is broken there is no fixing that. Probably a 7.0.
-
That big crease on the back really hurts it. Gonna have to go with the 7.0 people.
-
-
Darn, I got here late. At first I thought the second was the 9.8 as well, but then I saw the back upper spine corner has a slight chip. I’m betting that’s why they dinged it.
-
Man, that Gary guy just had to write his name on the cover of all his books.
-
-
-
The defects won’t really be helped by pressing. Lots of color breaking spine stress lines and then color breaking creases and then the chipping and tears. It’s not worth restoring. A 9.8 unrestored brings around $2500 and professional restoration services start at $1000.
-
Like the other guys said, get us a high res set of pics. From the low res pic you posted it looks like water damage which would severely impact the grade. Can’t be sure though without a higher res version.
-
Lots of color breaking defects so it's unlikely that you can improve it with pressing. I don't think I would try pressing either, because of that top staple. Probably 3.5.
-
Yep, 0% chance of an upgrade and likely it goes to a lower grade with those corners. Mine got graded recently and the only issue with it is that it has very light spine creases, the kind you have to angle the book in the right lighting to see. It got a 9.4.
-
-
Thanks folks. About what I expected. For $11 I can't complain too much.
-
3.0
I feel like people are being really optimistic given that water staining. That's a killer on grade usually.
-
- FoggyNelson and Galen130
- 2
-
4 hours ago, bc said:
While that sounds great - how about the interior defects (like pages & stamps missing, centerfold & wraps detached, tears, etc.)?
-bc
For my purposes experimenting with this sort of thing those won’t be accounted for, but you could take imagery of the interior of the book and model on that in addition to exterior imagery as long as the inputs are fairly consistent. The machine is just modeling off of what you give it. My initial stab at training the model will probably just be front and back cover images pasted together into one image with a grade attached. This is a fairly well explored category of machine learning starting way back when with facial recognition and the tooling is such that I can pick up a few libraries and get something going quick once I have a decent sized dataset composed.
All that said though it’s interesting to think of what you could do with high res images of books. If you’ve ever zoomed in on a 600dpi or 1200dpi scan of a cover you can see microscratches that are generally invisible to the human eye without a magnifying glass or microscope. These are the sorts of things that a computer will “see” quite easily. These are also the sort of things that can’t be corrected by pressing or other non-restoration techniques. This is where you could really go down a rabbit hole on grading. 9.991 for microscratch on cover. Lol
-
I'm thinking that I could create grading software as a machine learning project. I will need a lot of front and back images of comics that have been CGC graded (need the actual grade number as well) to train the model though.
-
3 hours ago, KCOComics said:
For some reason when I zoom in on the front cover it gets really blurry.
It looks really sharp from what I can see, but don't want to take a guess without being able to see the FC a little better.
It's a thumbnail rather than a downsized high res image. He needs to repost that one.
-
If I were to establish a systematic grading process here is what I would do.
- Assume that we use the same 10 point scale that is currently used (though IMO it's somewhat absurd and overly complex).
- Create a master document identifying each possible defect a comic can have and how it affects the grade of the book.
- Have five different people take the master document and the same exact sampling of books and give them all grades with notes on why they deducted points. This is done individually and in private.
-
Bring all of those people together and see where the differences occur.
- Are there things that can be interpreted differently in the master document?
- Are there errors in the master document?
- Are there things that are just wrong in the document?
- Are there things missing in the document?
Repeat the process until all you have is human error on the part of the graders, which you can't eliminate, but you mitigate by having multiple people grade a book.
Now that I think about it, this process might be a good match for a machine learning AI. You collect high res images of different books and their resulting grades and throw those into a model and you could probably get some pretty accurate scores. It would be impractical though as most imagery of comics out there varies pretty widely in terms of quality.
pgm Iron Man 128 ¡GRADED!
in Hey buddy, can you spare a grade?
Posted
It's not a bad looking book, but since we are supposed to be picky.
6.0