HGA grading

[leveljumper]

Originally Posted by rfgilles

To me, this is absolutely the most important question. They don't provide even the most basic info on their grading process, eg, how much is human grader vs software, weighing scores, cut-offs, is the software fully developed. If they are going to have any long-term credibility this info needs to be provided. To not disclose this information seems disingenuous.

Even their name "hybrid" suggests combining software with humans.

It seems their main selling point at this time is that they have the nicest looking slabs.

If they had fully software-driven objective grading and specified their criteria for cut-offs, I would definitely be interested.

I believe I’ve read in the past that they have a computer generated grade for centering and then a human grades the edges, corners, and surface. I could be wrong on that, though.


Sent from my iPhone using Tapatalk

[KhalDrogo]

Originally Posted by leveljumper

I believe I’ve read in the past that they have a computer generated grade for centering and then a human grades the edges, corners, and surface. I could be wrong on that, though.


Sent from my iPhone using Tapatalk

They have stated that their AI sends out errors which are then confirmed by two human graders, who also assess the cards for additional issues.

[rfgilles]

Originally Posted by KhalDrogo

They have stated that their AI sends out errors which are then confirmed by two human graders, who also assess the cards for additional issues.

That would seem reasonable if they do it that way. I'm still getting a Theranos, Elizabeth Holmes vibe. I hope they provide some proof that they actually use "AI", besides scanning cards and quantifying centering, which anybody with a scanner could do.

2 human graders seems like overkill if they are using "AI" as well.

[JWBlue]

Originally Posted by SupermanBrandon

I've said it before and I'll keep saying it...you come into a market desperately seeking transparency. You claim you are changing the game w/ AI technology. Yet the only thing we've seen are "sick, dope" labels. Marketing 101 says you lead w/ your big 900lb gorilla...the AI. Show it off. Show it working. Give social content around it. Yet, we have nothing. Barely a mention of it. Then you have errors of mis-spelled names, wrong labels, subs not adding correctly, and encasing errors. As others have said...you can't even get your computers talking to each other to print the right labels. How in the world did you build game changing AI grading technology?

I was going to send cards to them until I read this. Seems accurate.

[uberfractor]

Originally Posted by rngrdanny22

Anyone figured out, or heard, what type of "AI" they are using?

Or is it just a marketing ploy?

Best I can tell, this is just GMA in a pretty package. How many employees do they have? Are they also located in a basement?

Quote from their website (emphasis mine):

The cards should be graded solely on the presentation of the card itself. So, we are developing unique software that detects edges for crispness, corners for sharpness, centering for balance, and surfaces free of blemishes; we feel that we will grade your card accurately 100% of the time.

This sounds like they don't actually have this technology. I'm guessing this means they sat around a table and agreed this software was a good idea and now they are going to pay some hourly programmer to implement some open source image processing library to look at scans. There's a reason why none of us have seen this technology actually working. That's because it's not working. My guess is they are pretending to being doing this (or doing it with janky terrible software that's basically pointless) with the plan of using revenue to improve the tech over time. I don't think they are doing it maliciously, they probably think they are doing some good by trying to do this. I don't disagree with it being a good concept but executing is going to be very difficult.

Also, I don't mind the design of the slabs but it looks like it was printed on cheap paper and laminated by a $40 machine from amazon. I can see tons of spots, I think they are running their laminator too cold. They could at least make it look good when printed by getting a higher end printing setup.

[tmoore_25]

I can't find it because the search function on here is horrible or maybe I'm just not doing something right, but I remember a while back there was a member who was developing some sort of AI grading technology. He solicited info and advice from some of the forum members here.

Is this the same guy the runs HGA?

[rfgilles]

Originally Posted by uberfractor

Quote from their website (emphasis mine):



This sounds like they don't actually have this technology. I'm guessing this means they sat around a table and agreed this software was a good idea and now they are going to pay some hourly programmer to implement some open source image processing library to look at scans. There's a reason why none of us have seen this technology actually working. That's because it's not working. My guess is they are pretending to being doing this (or doing it with janky terrible software that's basically pointless) with the plan of using revenue to improve the tech over time. I don't think they are doing it maliciously, they probably think they are doing some good by trying to do this. I don't disagree with it being a good concept but executing is going to be very difficult.

Exactly. If they are currently relying on human graders, they are no different from any other grading company. They are being misleading at best, and fraudulent at worst. Not credible either way until they provide some evidence that they are using software for grading.

[rfgilles]

Just posted on Facebook. Doesn't move the needle for me until they provide videos or pics of how the software works.

New advances in algorithm and hardware design have supported the emergence of artificial intelligence technologies in applications ranging from self-driving cars to cancer detection at superhuman rates. At HGA we are bringing those innovations to the sports card community. These new technologies work with the inherently objective and quantitative nature of machine grading to usher in a bright future where all cards are graded more fairly and reliably than ever before.
The grading process begins with a cleaning step before a high resolution scan is taken of each face of the card at 3200 DPI. That’s around the equivalent of 90 pictures captured on your smartphone. The high resolution scan is then run through our state-of-the-art automated grading system built from our vast database of annotated image data. Using various image transformations, convolutional operations, and other machine vision techniques, it’s able to extract the key features and measurements from the card image. Our predictive model then computes scores for various card component grades based on the extracted information. We are constantly working to improve our model and to stay on top of all of the new designs and trends from card makers. Once the machine grading is complete, we use two expert human graders to double check the work and ensure our high standards are always met.

[mooselang]

Originally Posted by rfgilles

Just posted on Facebook. Doesn't move the needle for me until they provide videos or pics of how the software works.

New advances in algorithm and hardware design have supported the emergence of artificial intelligence technologies in applications ranging from self-driving cars to cancer detection at superhuman rates. At HGA we are bringing those innovations to the sports card community. These new technologies work with the inherently objective and quantitative nature of machine grading to usher in a bright future where all cards are graded more fairly and reliably than ever before.
The grading process begins with a cleaning step before a high resolution scan is taken of each face of the card at 3200 DPI. That’s around the equivalent of 90 pictures captured on your smartphone. The high resolution scan is then run through our state-of-the-art automated grading system built from our vast database of annotated image data. Using various image transformations, convolutional operations, and other machine vision techniques, it’s able to extract the key features and measurements from the card image. Our predictive model then computes scores for various card component grades based on the extracted information. We are constantly working to improve our model and to stay on top of all of the new designs and trends from card makers. Once the machine grading is complete, we use two expert human graders to double check the work and ensure our high standards are always met.

This is all I could think about when reading that post:

[Bcr]

Originally Posted by mooselang

This is all I could think about when reading that post:

Haha. That image is exactly right.

[salthill]

Originally Posted by rfgilles

New advances in algorithm and hardware design have supported the emergence of artificial intelligence technologies in applications ranging from self-driving cars to cancer detection at superhuman rates. At HGA we are bringing those innovations to the sports card community. These new technologies work with the inherently objective and quantitative nature of machine grading to usher in a bright future where all cards are graded more fairly and reliably than ever before.

There’s a whole lot of waffle in that pitch.
Including ignoring any bias in the training data. (ie it’s not “inherently objective”.)
Also, they’re conflating artificial intelligence and machine learning.
They could be legit, but until they prove otherwise, I’ll remain unconvinced.
(Remember, it’s on them to prove their tech works, not us to disprove it.)

[leveljumper]

Originally Posted by salthill

There’s a whole lot of waffle in that pitch.
Including ignoring any bias in the training data. (ie it’s not “inherently objective”.)
Also, they’re conflating artificial intelligence and machine learning.
They could be legit, but until they prove otherwise, I’ll remain unconvinced.
(Remember, it’s on them to prove their tech works, not us to disprove it.)

I agree with you they are conflating machine learning with AI, but this is common in society in general and I view the phrasing more as marketing speak. There are a lot of people that talk about things having “AI” that are not AI.


Sent from my iPhone using Tapatalk

[salthill]

Originally Posted by leveljumper

I agree with you they are conflating machine learning with AI, but this is common in society in general and I view the phrasing more as marketing speak. There are a lot of people that talk about things having “AI” that are not AI.


Sent from my iPhone using Tapatalk

True. I’d be much more concerned with what their training data is.
If they’ve just used a bunch of images of PSA and BGS graded cards, then congrats, they’ve just built a system that - at its best - can only equal the performance of PSA/BGS.

[leveljumper]

Originally Posted by salthill

True. I’d be much more concerned with what their training data is.
If they’ve just used a bunch of images of PSA and BGS graded cards, then congrats, they’ve just built a system that - at its best - can only equal the performance of PSA/BGS.

I think that’s their goal, to use the PSA or BGS standard but to just be consistent with it and take market share that way. That’s a big complaint a lot of people have with BGS and PSA is that there isn’t consistency with what’s mint vs gem mint, etc. Being a “tough grade” doesn’t mean anything. A lot of people consider SGC the toughest grade, what has that done for resale value? Resale value is the point at the end of the day. Like 90% of people grade with the primary goal being to increase value of cards, not PC slabs. I personally think they probably should have done 3 different 10 grades (gem mint, pristine, flawless) instead of the BGS route 9.5 gem mint with 2 10’s and think it will be a huge mistake at the end of the day. As of now they’re doing great, though, and I’m looking forward to getting my cards back from them.


Sent from my iPhone using Tapatalk

[PepperDean]

There are so many issues with this.

(1) If this automated grading process is so superior to the grading processes offered by the other grading services, why would they not have pursued "inventing" it? They certainly have the capital to do so. Presumably, not only is it more accurate, but also simply being able to insert a card into a machine and having it then produce a grade should dramatically expedite the process as well, which would certainly appeal to the giants like PSA and BGS.

(2) And this brings me to my second point. If HGA has this new, automated, fast grading system, then why does it have such a hard cap limit on the number of cards submitted per week, and in particular, why is this number so low? Even as a new company with a small staff, if the grading software works as they advertise they should be able to grade far more cards. The current speed of grading they advertise implies a grading system similar to if not identical to those offered by the other companies, which means they are probably grading by hand.

(3) If they have this automated grading system, of which the calling card is "100% accuracy," why then have we already seen examples where cards have folded corners or are deeply off centered getting Gem Mint grades? Does the software not work? Because, if it does, it shouldn't just rarely make mistakes - it should never make them. Or, are they actually just grading by hand, like I stated before? Also, I understand having people check grades after they're assigned as a safety net, but if they stand by their software this really shouldn't be necessary. The inclusion of that step seems to me like a subtle admission humans ultimately grade the cards.

(4) The only bit of their "automated process" they've been explicit about is the photo taking. So, is their automated process just having a nice camera? Seems like it.

(5) There's not such thing as purely unbiased grading, as they advertise. Even the code that would evaluate cards in such an automated system would have to be written by a person, which creates bias around certain card features being evaluated more strictly or weighed more heavily than others (towards the final grade). Now, if this code does exist it should created unbiased grading from card to card, but that doesn't mean bias doesn't exist in the process at any step. Also, cards don't wear and become damaged in uniform stages. There is no such thing as the ideal "10", "9", "8" because of this inescapable reality. There always needs to be a judgement.

(6) We saw CSG grading, from an already massive company with established business standards and grading experience across collectibles, announce they would enter the market and then finally do so half a year or more later. But HGA? Where was the longstanding announcement of entry? How did they come up with this "automated system" so quickly, and just jump into the game almost overnight? How on earth were they able to do this when it took CSG, with all the pre-existing infrastructure, so long to get started? And who from the industry did they bring in to learn how to grade or write an appropriate grading code? What qualifies them to assign a grade? Sounds like some guys saw the market needed more grading, and just jumped in quickly to make a quick buck.

(7) The language they use on their website sounds like it was written by a "get rich quick" clout-chasing college bro who spends more time on Tiktok than committed to bettering his business. "Customer Service is the best, bar none," with an incorrectly capitalized second word. "Easy to submit" with no punctuation after it, unlike every other bullet point on their home page (just unprofessional inconsistency). "HGA is revolutionizing the industry" using buzzwords to attract the easily persuaded and gullible. "Changing the game," a literal phrase I've heard coming out of every college bro's mouth (as someone who not long ago was in college). Is using this language inherently problematic? No. But compounded with the other data points, like constant misspellings and improper grammar, and an obvious focus on appearance rather than on the quality of their service, it reeks of young, inexperienced, and quite frankly, a scam.

To put it bluntly, it seems like HGA is just the next "get rich quick" scheme concocted by a couple guys as they followed the rep around in Target on their last one.

[Oldan Poar]

. .

[ArmyDC]

I would trust CGS 100x more than HGA, they have been grading stuff for a pretty long time.

[rfgilles]

Originally Posted by PepperDean

There are so many issues with this.

(1) If this automated grading process is so superior to the grading processes offered by the other grading services, why would they not have pursued "inventing" it? They certainly have the capital to do so. Presumably, not only is it more accurate, but also simply being able to insert a card into a machine and having it then produce a grade should dramatically expedite the process as well, which would certainly appeal to the giants like PSA and BGS.

(2) And this brings me to my second point. If HGA has this new, automated, fast grading system, then why does it have such a hard cap limit on the number of cards submitted per week, and in particular, why is this number so low? Even as a new company with a small staff, if the grading software works as they advertise they should be able to grade far more cards. The current speed of grading they advertise implies a grading system similar to if not identical to those offered by the other companies, which means they are probably grading by hand.

(3) If they have this automated grading system, of which the calling card is "100% accuracy," why then have we already seen examples where cards have folded corners or are deeply off centered getting Gem Mint grades? Does the software not work? Because, if it does, it shouldn't just rarely make mistakes - it should never make them. Or, are they actually just grading by hand, like I stated before? Also, I understand having people check grades after they're assigned as a safety net, but if they stand by their software this really shouldn't be necessary. The inclusion of that step seems to me like a subtle admission humans ultimately grade the cards.

(4) The only bit of their "automated process" they've been explicit about is the photo taking. So, is their automated process just having a nice camera? Seems like it.

(5) There's not such thing as purely unbiased grading, as they advertise. Even the code that would evaluate cards in such an automated system would have to be written by a person, which creates bias around certain card features being evaluated more strictly or weighed more heavily than others (towards the final grade). Now, if this code does exist it should created unbiased grading from card to card, but that doesn't mean bias doesn't exist in the process at any step. Also, cards don't wear and become damaged in uniform stages. There is no such thing as the ideal "10", "9", "8" because of this inescapable reality. There always needs to be a judgement.

(6) We saw CSG grading, from an already massive company with established business standards and grading experience across collectibles, announce they would enter the market and then finally do so half a year or more later. But HGA? Where was the longstanding announcement of entry? How did they come up with this "automated system" so quickly, and just jump into the game almost overnight? How on earth were they able to do this when it took CSG, with all the pre-existing infrastructure, so long to get started? And who from the industry did they bring in to learn how to grade or write an appropriate grading code? What qualifies them to assign a grade? Sounds like some guys saw the market needed more grading, and just jumped in quickly to make a quick buck.

(7) The language they use on their website sounds like it was written by a "get rich quick" clout-chasing college bro who spends more time on Tiktok than committed to bettering his business. "Customer Service is the best, bar none," with an incorrectly capitalized second word. "Easy to submit" with no punctuation after it, unlike every other bullet point on their home page (just unprofessional inconsistency). "HGA is revolutionizing the industry" using buzzwords to attract the easily persuaded and gullible. "Changing the game," a literal phrase I've heard coming out of every college bro's mouth (as someone who not long ago was in college). Is using this language inherently problematic? No. But compounded with the other data points, like constant misspellings and improper grammar, and an obvious focus on appearance rather than on the quality of their service, it reeks of young, inexperienced, and quite frankly, a scam.

To put it bluntly, it seems like HGA was just the next "get rich quick" scheme concocted by a couple guys as they followed the rep around in Target on their last one.

Agreed! They claim to have not 1, but 2 graders check the software results. Doesn't seem feasible.

[HalfNipponese]

I have some rare vintage cards that are probably going to be sold through Goldin in the near future. Right now I'm leaning towards CSG or HGA just to see how the market will react. I'm hoping they will accept HGA and give AI a chance to show how it'll be accepted by the vintage collector community. If Goldin gives me the OK the cards will be off to HGA with CSG being my second choice.

Of course I'll listen to the professionals if they tell me to only use PSA.

[kabrune2]

Originally Posted by HalfNipponese

I have some rare vintage cards that are probably going to be sold through Goldin in the near future. Right now I'm leaning towards CSG or HGA just to see how the market will react. I'm hoping they will accept HGA and give AI a chance to show how it'll be accepted by the vintage collector community. If Goldin gives me the OK the cards will be off to HGA with CSG being my second choice.

Of course I'll listen to the professionals if they tell me to only use PSA.

If your cards are high enough in value to be put in Goldin I’d seriously consider PSA as your 1st, 2nd, and 3rd choice, unless there’s a dent that PSA would grade no higher than a PSA 6, but another company might give an 8.

But I’d be interested to see how an HGA slab does in a bigger auction setting.

[i4gotmyid]

Originally Posted by kabrune2

If your cards are high enough in value to be put in Goldin I’d seriously consider PSA as your 1st, 2nd, and 3rd choice, unless there’s a dent that PSA would grade no higher than a PSA 6, but another company might give an 8.

But I’d be interested to see how an HGA slab does in a bigger auction setting.

Or...please do grade through HGA, I wouldn't mind trying to get some nice cards at a discount

[HalfNipponese]

Originally Posted by kabrune2

If your cards are high enough in value to be put in Goldin I’d seriously consider PSA as your 1st, 2nd, and 3rd choice, unless there’s a dent that PSA would grade no higher than a PSA 6, but another company might give an 8.

But I’d be interested to see how an HGA slab does in a bigger auction setting.

I know that I'll make more with PSA but I'm just so curious. I appreciate the advise as this will be my first time selling a vintage card. The three cards that Goldin are interested in are:

1976 Sport Parade Americana Jimmy Connors Rookie-PSA Pop 2
1976 Sport Parade Americana Martina Navratilova Rookie-PSA Pop 0
1976 Sport Parade Americana Chris Evert-PSA Pop 3

They will all probably grade between a 5 and 6.

[rohara99]

I have no plans to ever submit to HGA (so I have no skin in the game) but why do people keep referencing the Mark McGwire card with the Gem Mint folded corner?

It was damaged while being slabbed and HGA owned up to it and has bought it back at comparable BGS 9.5 value. Thats more then any of the big 3 would do if they made a similar mistake.

[SupermanBrandon]

Originally Posted by HalfNipponese

I have some rare vintage cards that are probably going to be sold through Goldin in the near future. Right now I'm leaning towards CSG or HGA just to see how the market will react. I'm hoping they will accept HGA and give AI a chance to show how it'll be accepted by the vintage collector community. If Goldin gives me the OK the cards will be off to HGA with CSG being my second choice.

Of course I'll listen to the professionals if they tell me to only use PSA.

Do you love wasting time and money, by chance?

[DR Murph]

I think many are underestimating the potential of HGA.