JC25427N

You forgot to add "Which we didn't have to do...."

Is your name Bree by any chance?

Both of you nerds need to quiet down, also this is a profit/returns message board* get your facts straight

Because of their lack of understanding

Did you rent those pom-poms or do you own them

That's a beauty, the person who submitted it to you definitely doesn't lack understanding!

Am I too late to preorder a copy of the 2024 edition

All good Is it still going to be 5 total submissions across all categories per person again?

Will DeLoreans be provided or is it BYOTM (Bring Your Own Time Machines)

It was @greggy

There's a bit of a workaround that I mentioned back around page ~70 or so, you still can't look up past IDs, but if you want to verify whether an ID used to belong to someone else you can check their feedback page: https://www.ebay.com/fdbk/feedback_profile/aureliaag (There's a symbol by their name that says that this person's ID has changed recently which somewhat helps) and then replace the ID in that url with the ID you suspect might be a previous ID of that person https://www.ebay.com/fdbk/feedback_profile/zaneglor If the feedbacks listed on both pages are the same, then it's the same person. Wishing everyone a Merry Christmas, don't forget to spend some time with your loved ones, this thread will still be here after the holidays don't worry

We solved that some tens of pages ago, the Zaneglor account was never kicked off, he just changed the id of that account to aureliaag, hasn't sold a book on that account since ~2020

Well wasn't it mentioned before, that in the slab model used just before the change in 2016, the label was actually attached as part of the inner well. So with our current running theory of how this scam is working (replacing the inner wells of slabs), I don't think it would have been possible to do this exact exploit before the the slab iteration post July 2016.

nevermind, I'm not gonna try anymore, I don't think I'll get anywhere. Good luck

I'm not sure if you're lumping me in there, so just to be clear, I'm not against it by any means, I work in software development, I just don't think there's an appreciation for how complex it would be to create. I also think some over-estimating might be going on, processing large image datasets can take a really long time, might not be as fast and efficient as people think.

It's more of a "can", not "will". It's a possibility it will be regraded, not a certainty.

I think he's talking about the issue being discussed in tl;dr so you don't have to shift through 70 pages is that that seller, briva3, has basically been all but confirmed to be pulling a scam where he is swapping lower grade books into high grade label slabs and reholdering them.

Alright, I could believe that, but are all those scans of sufficient quality? You wouldn't really know until you train the model and look at the training results I suppose. One thing that could be a source of noise is if you have scans of differing quality in the dataset, that could introduce bias in the model. There's a whole lot of dimensionality to consider here, I'm just going off the cuff with what I would be thinking if I were asked to do this. It would take a lot of work, I'm not saying its impossible, but its not as easy as just saying "train the computer to do this" edit: Sorry I just read all the posts in this thread asking to stop the AI talk, this thread moves at lightning pace forgive me I just saw it

I think there's a slight confusion here I think when you say scanning, you mean looking at an image and interpreting it, drawing conclusions from it, etc. Yeah I'm sure AI could do that for comic books if all preconditions are met for training such a model. When I say scanning, I mean taking a picture (taking a scan). So the concern raised by some CG expert earlier is that the picture taking technology in the current state is not sensitive enough for training a model to accurately grade. I don't believe it either personally (my main issue in that rant I went on was about data collection with the assumption all necessary machinery exists), but I was just repeating what someone else who claims to be a subject matter expert on CG (which I am not, so I kind of defer to them) said earlier .

Yeah this is one concern you would have when trying to train this AI, this is part of data integrity, insuring that the data doesn't have any noise (such as dust particles) that mislead the model, creates error, creates bias, etc.

Ok, what I'm talking about (and what I thought you were talking about) is a machine learning model (AI) that grades comic books (perfectly, accurately, fast, and without bias as you mentioned before). To give you just ONE of many aspects that make this harder than someone who doesn't work with this kind of stuff may think, you'd have to collect enough data to train this hypothetical model (I'm not even talking about any physical robotics of flipping pages, scanning, turning, etc that VintageComics and everyone else seems to be concerned with, lets just say all of that machinery exists). To even begin the conception of creating a model you need to collect a dataset to train the model on. To do this you would have to either collect a large corpus of graded books (as you said you want this model to be trained only on books that have been seen and graded before), crack them out of their slabs, and then collect all the necessary scans that you would need to grade it, or alternatively. get a large corpus of books, have someone at CGC grade them, and take all the necessary scans before encapsulating them (and this wouldn't just be the front cover and back cover scan that CGC provides now, if you want this model to do full grading end to end you need a scan of every page of every comic book that CGC could ever grade to assure that the book is whole and correct without incorrect married pages, missing inserts, etc). Then you probably need multiple examples of each book at different conditions. Then you have to know that whatever form of input data is used to train the model will determine what you actually need to give the model when it's used in production for actual grading, so if you give it 50 scans per book for training, you need to collect those scans in production. Collecting this data and collecting it at the quality necessary for good performance (which according to that one person (sorry I forgot the name) who posted earlier who said they have a vast experience in CG and working with scanning technology says that even the most sensitive current scanners wouldn't be sufficient for this task, I don't personally know that, just repeating it). So collecting this data is a herculean task, making sure the data integrity and quality is sufficient is a herculean task, and even storing all of these hi-res scans (again, every page, of every book, at multiple angles, and you might need to repeat this for multiple conditions) would get insanely costly, and I don't even mean just storing the training data, you would probably want to store the scans used in production in case something goes wrong for validation, someone wants to appeal a grade, etc (and then go back and see how many submissions CGC gets). And this is just data collection, one of the first steps when deciding to create this model, past that there are other hurdles to jump through. Again just for the machine learning model that goes the grading, I've assumed that the scanning machinery exists (although apparently thats up for debate whether current scanning capabilities are enough, I would imagine it is but I'm not a grader or a CG expert)

This is insanely harder than you make it sound