Vision based Card Game with Jetson TX1/TX2 + USB WebCamera

I’ve been exploring the feasibility of using an NVidia TX1/TX2 to play live board/card games alongside a human companion.

As part of this exploration I created a detectnet model to detect playing cards, and a separate ImageNet model to detect the face card type. Then I used those two datasets as part of a very simplified game of blackjack.

If you want to check it out I created a fork of the official jetson-interference repository, and added my changes to it.

At this point I’m not sure what else I’ll add to it. I’m going to try out segmentation networks to solve the issue of not being able to stack cards on top of each other. I’m also going to explore reinforcement learning to add decision making to it.

I’m also hoping that other people can add stuff to it because overall I think the Jetson TX1/TX2 has a lot of potential with this kind of application. I’m also not much of a coder, and most of what I do is hacks onto existing code.

Cool project, thanks for sharing! I will add it to the list of example projects on the wiki if you don’t mind. Novel use of detectNet and imageNet pipelined together, maybe even segNet too in the future.

If you are interested in trying your hand at reinforcement learning, pyTorch seems to work pretty good on Jetson TX1/TX2 and comes with RL examples: https://gist.github.com/dusty-nv/ef2b372301c00c0a9d3203e42fd83426