Our short welcome video will introduce you to the event and how to get started
In order to play our Custom Vision Rummy game, you need to train your own image classification AI to be able to identify planets in images. We'll use Microsoft's Custom Vision service to help us create and train an AI image classification model. When you have done this, you can use your newly created model to play our Rummy Vision game and compete in the tournaments for our event.
To proceed, work your way through the instructions on this page. Start off by:
Signing up for an account on the aigaming.com site.
Creating a Microsoft Azure account if you don't already have one.
Your event host will help you create an account, or you can follow our instructions at Create a Student Free Trial Microsoft Azure Account.
Then, continue working through this page, watching the videos below for the next steps and for detailed instructions of how to
train your own Custom Vision Model
access that model from Python code
use it to play our Rummy Vision Planets game
Microsoft's Computer Vision service is a collection of pre-trained AI models that recognise objects in images like animal, landmarks, celebrities and words. Our game needs you to be able to identify the different planets which Microsoft's Computer Vision does not do, but, you can create and train your own Microsoft Custom Vision model to do this. The next video shows you how to create and train your own Custom Vision model to recognise planets.
You can download the initial set of training and test images that are referred to in the video above from:
Or you can see it in our GitHub repo at:
This video shows you how to call your new Custom Vision model from Python in order to analyse images of planets. It tells you where to find our template code that will show you how to add your Custom Vision model to your code and it lets you analyse your own planet images to see how well your Custom Vision model performs.
Now that you can call your Custom Vision model from a code file you can experiment with other planet images and with how you use the Custom Vision model.
Now that you have seen how to call your Custom Vision model to analyse images, we can use the model to play the Rummy Vision game. We've got some more template code to get you started. Watch the video below to find out how to use it with your Custom Vision prediction url and prediction key.
You've now seen all of the basics to be able to play our Rummy Vision game with planet images. We recommend you concentrate on improving you Computer Vision model by training it with more images.
If you want to find out more about how to improve the template code that plays the Rummy Vision game, make sure to watch the next video Introduction to coding your game playing bot
Now that you have been to able to create and train your own specialised Custom Vision model to detect planets, you can use it to compete in our Rummy Vision game.
Learn exactly where you need to add your code to develop your game playing bot. What information you will receive, and what information you need to return in order to play the game.
Entering your code into a tournament lets you find out how good your game playing bot is. Each event will have at least one tournament and the video below gives you a quick overview of how to make sure your code is registered to play.
The Online Code Editor is where you spend most of your time as you write the code for your automated game playing bot. It's also where you run the code to play the games. Find out all about the Online Code Editor and how to play games in this video
JSON objects are widely used to transfer data to and from API services. They are human readable text strings which adhere to a formal syntax which means they are also readable in software. JSON objects can be easily manipulated in Python code by converting them to or from dictionary objects. This video introduces the format of JSON objects, demonstrates how to convert them to and from Python dictionary objects, and gives examples of how to work with dictionaries in your code.
We recommend tackling the following steps in order as the best approach to improving your code:
Increase the number of images for each planet to improve the accuracy of your model. Use at least 20 images. More if you can.
Test your model in the ComputerVision.ai site dashboard to gauge which planets it struggles to recognise and concentrate on improving those planets.
You can read more detail about training your Custom Vision model at https://docs.microsoft.com/en-us/azure/cognitive-services/custom-vision-service/getting-started-improving-your-classifier
If your model is successfully recognising planets, you can look at adding some game playing strategy to the code that plays the Rummy Vision game on the AI Gaming site.
You can read more about improving the Rummy Vision code on our Microsoft Rummy Vision Help Page
For more detailed information about developing your solution for the Rummy Vision Game, go to the Microsoft Rummy Vision.
For our written Really Quick Start summary of how to starting coding solutions on the site go to our Really Quick Start page here.