It's pretty common these days to hear words such as: Artificial Intelligence (AI), machine learning, computer vision, etc. I'm sure that you hear or read at least one of those words once a week right? I won't give you all the technical details of this, but, I will give you some information so you know better what to expect.

One challenge in listing cards online as you already know, is filling in all the information so that a user can search and find your card to buy it from a marketplace. Filling in all the information has also been a challenge for us as collectors, so we understand the process. We tried to automatically identify cards in 2017 using some image recognizing techniques but these were not successful to the degree we would have liked. So we decided not to continue with that feature. But since then, there has not been a week or day that passes without us thinking about the dream of having a Kronocard that can document your card for you. As we kept reading and learning new stuff about artificial intelligence every month, one day something popped into our head. Instead of trying to match two card images together to identify the card, why not try to extract as many information as possible from the card and try to figure out what it is. We did this by training a computer to recognize the characteristics of a collectible card! Eureka!

That's were it started, in January 2020 we started to assemble a team to tackle this near impossible task. As you know, sports cards particularly, are very different from each other in design, giving computer learning lots of challenges. After 18 months, the same incredible team has been able to get unexpectedly good results. Our target was that this AI documentation would be able to correctly identify around 40-60% of the cards. However we pushed the system so hard that it surpassed our expectations and we have been able to achieve a 75-95% detection rate! This is an out of this world achievement for such a small team with limited budget and time.

OK so what's the catch? Why between 75-95% ? Why such a large discrepancy?

Here's what to expect:
It will not document 100% of your cards, period!

If this is what you’re expecting, well, you'll be disappointed, sorry.

Here are some more details about the process and why it cannot achieve 100% accuracy.


Because we needed to train a computer to understand the structure and data that is found on each type of card, we concentrated our effort on the major types first.

  • Baseball
  • Football
  • Basketball
  • Hockey
  • Soccer
  • Pokémon
  • Magic the Gathering

We will add support for other types of cards in the future but before we do that, we will first fine tune those seven. If you try to submit other types of cards, Kronocard will not send it to the AI recognition server.

Also very important to remember, for the moment we can't identify cards that have been released in 2021. We need to train the system to be able to recognize those new cards but with all the work to be done we are a little behind with the supported cards. Don't worry, this is a priority and as soon as the system is released officially we will catch up. Then we will put a team in place to train newly released sets as soon as possible.


Something important to understand, because of the nature of the task and the hardware required to analyse a card, it's not possible for Kronocard to have this integrated and working in real-time directly on your computer. To document cards, you will still have to submit them to our documentation server but this time without filling any information. Scan and submit! As soon as the request is received, 24/7 365 days per year, the system will work on your card. At the present moment the current capacity of our AI servers is around 500-800 cards per hour for sports cards and 1,000-1,200 for TCG. But, don't necessarily think that by submitting 200 cards now you will receive them in maximum one hour. This is the total server capacity, so if other users submit cards before you, it's a first in first out process. But, don't worry we will monitor the system and add more computing power when needed.


Because of the nature of sports cards, manufacturers and designers are getting more and more creative in providing different designs for the market. Up to the 1900's, most, if not all card design, was done in an old-school fashion by hand and without a computer. Since the computer era, we have started to see a lot more funky and interesting designs in sports cards. And in the more recent years, many more printing techniques have appeared helping to create many more interesting variations and looks.

That being said, for our AI documentation, the more basic the card design the more chances that the card will be correctly identified. Wait! That doesn't means that you can only submit cards from before the year 2000. We have finetuned our system to also work with many complex designs, but there is a limit to this, you must understand.

Here, what is the difference between those two cards?

The one on the left is the regular version, the one on the right is the blank back version... Second challenge for a blank back version, with only a name on the front and without any other information, the AI system will probably not be able to identify that card. Don't forget that our system is extracting the information on the card, it is not comparing image pixels.

Another example of the possible problems while identifying your cards. There are tons of situations like this applicable to sports cards. Look at the four images closely, they are all different! (1st down, 4th down, base, kickoff) Ho! There are also other versions of the same card! Touchdown, Goal Line, 3rd Down, Red Zone and probably even more!

In such cases what will be returned? The AI in most cases will try to return as much information as possible without sending back false information (not always the case). If the AI is confused about all the possible names to return for the set. It will return only what is common between all results. In this case we do have this:
Panini Playoff 1st Down
Panini Playoff 4th Down
Panini Playoff
Panini Playoff Kickoff

So the name of the set returned will be: Panini Playoff.

Don't forget that we have based our system on data found on the cards; logos, text, signatures, serial numbers, etc. We also started to incorporate color variation detection but there are lots of challenges in color detection. That means our system can't really see a difference between those two cards. By chance, maybe it will see that the one on the left has a serial number, have you found it?

Same here, can you see the difference? Our system can't too... The one on the left is the gold version, the one the right is the orange version. Starting to understand the challenges now? And I didn't talk about the quality of the player name at the bottom of the card. 😂

And now speaking of challenges in the method the design is done. The computer needs to find where the player name is, not an easy task. Although you're a human with eyes, the computer only sees 01010011101100101 you know?

Same here for Pokémon, only based on the image, text and logos we can't really see a difference other than the background that our system cannot really see.


So, I could have listed thousands of examples like this but I think you will now begin to understand why our system cannot return the right information all the time while documenting your cards. It really depends on the design, we have been able to get 100% accuracy on 800 cards at some point because the designs were clean and simple. We really wanted to take our time and explain how this AI documentation works before you subscribe to it. We don't want to fool you by promising that your cards will all be documented accurately, we want to be transparent with our users.

We hope that you will enjoy what we have made to help you through your journey of digitizing your entire card collection!