Pokémon TCG Search Engine: Use AI to Catch Them All

With a new custom image similarity service, we are able to build an image search engine for collectible cards trading.
avatar
Michal Lukac
16. June 2021
Collectible card trading game Pokémon.

Did you play any trading card games? As an elementary school student, I remember spending hundreds of hours playing Lord of the Rings TCG with my good friend. It was a fantastic game. Back then the LOTR was in the cinemas and the cards were simply beautiful. I played with a combination of Ents and Nazguls. Other people in our office spent their youth playing Magic The Gathering (yes, their card artworks are amazing as well), and collecting the sports cards with their favorite players. I personally liked the basketball cards. Cards are still loved, played, collected, and traded by geeks, collectors, and sports fans across the world!

In the last year, we have been building a system able to train visual models with numerous applications in image search engines. We already offer visual search services for photo search. But, they are optimized mostly for general and fashion images. In the last decades, we all witnessed the growth of the TCG cards market. However, technologies based on artificial intelligence have not yet been used in this market. And since it’s a perfect match, we are going to change itwith a card image search.

Contents:


Where can you use image search for cards?

Trading card games can have tens of thousands of images. In principle, building a basic image classifier leads to low precision and is simply not enough for more complicated problems. However, we are able to build a complex similarity system that can recognize, categorize, and find similar cards by a picture. Once we train it properly, it can deal with enormous databases of images it never encountered before. With this system, you can find all the information, such as the year of release, card title, exact value, card set, or whether it already is in someone’s collection, with just a smartphone image of the card. The collectibles are a big business and some cards are really expensive nowadays. Who knows, maybe you have the card of Charizard or Kobe Bryant hidden in your old box in the attic.

Which TCG cards could visual AI help with?

An image search engine is a great approach when the number of classes for the image classification is high (above 1.000+). With cards, each card is representing a unique class. A convolutional neural network (CNN) trained as a classifier can have poor results when working with a larger number of classes. Pokémon TCG contains more than 10.000 cards (classes), Magic the Gathering (MTG) over 50.000, and the same goes for basketball or any other sports cards. So basically, we can build a visual search system for both:

  • Trading card games (MTG, Lord of the Rings, Pokémon, Yu-gi-oh!, Warhammer, …)
  • Sport collector cards (Topps, Upper Deck, Panini, …) like Ice Hockey, Football, Soccer, Baseball, Basketball, UFC, and more.
Pokémon, Magic The Gathering, LOTR, Ice Hockey and Basketball cards.
Pokémon, Magic The Gathering, LOTR, Ice Hockey, and Basketball cards.
Yes, we are big fans of all these things 🙂

A similar image recognition technology is starting to be used on e-bay platform when listing trading and sports cards for sale. However, this is only available in the e-bay app on smartphones. The app has a built-in scanning tool for cards and can find the average price with additional info. Our service for card image search can be integrated into your website or application. And you can simply connect via API through a smartphone, computer, or sorting machine to find exact cards by photo, saving a lot of time and improving the user experience!

We’ve been recently training a CNN model for Pokémon cards image recognition. Why Pokémon? Pokémon is the most played TCG in the world, the game has simple rules and an enormous fan base. Some cards are really expensive, but more importantly, they are traded heavily!

With this model, we built a reverse search for finding the exact Pokémon card, which achieved 94%+ accuracy (i.e. exact image match). And we are still talking about a prototype developed in a few days that can improve to almost 100 %. Watch this short video with a test of finding Pokémon cards by image.

Example of our custom similarity search engine that was trained on Pokémon trading cards

With the new custom similarity service, we are able to create a custom solution for trading card image search in a matter of weeks. The process for developing it is quite simple:

  1. We will schedule a call and talk about your goals. We will agree on how we will obtain the training data that are necessary to train your custom machine learning model for search engine.
  2. Our machine learning specialists will assemble a testable image search collection and train a custom machine learning model for you in a matter of weeks.
  3. After meeting all the requirements of PoC, we will deploy the system to production, and you can connect to it via Rest API.

Machine learning models bring endless possibilities not only to pop culture geeks and collectors, but to all fields and industries. From personalized recommendations in custom fashion search engines to automatic detection of slight differences in surface materials, the visual AI gets better and smarter every day, making routine tasks a matter of milliseconds. That is one of the reasons why it is an unlimited resource of curiosity, challenges, and joy for us, including being geeks – professionally :). Any idea came to your mind? Let’s discuss it.

Related Articles

A step-by-step guide for using image similarity as a service. Find similar items with accurate & fast API for Image Search.
Read moreJuly 2021
Annotate is an advanced image annotation tool supporting complex taxonomies and teamwork on computer vision projects.
Read moreMay 2021
After years of experience in e-commerce, we developed the Fashion Search enabling sellers to create their own fashion product discovery systems.
Read moreApril 2021