Image registration and recognition model for the card game Set
The Set card game is awesome! However, sometimes it's tricky to check if a set exits on the board. The goal of this code is to make an image recognition project to do just that: to be able to take a photo of the board and output all possible sets.
Desired components
A brief rundown of what I want to build into this
Tasks
- Code to take in raw images of single cards and resize/pad
- Image augmentation to take a small number of photos of Set cards and generate more ad infinitum using imgaug library (which is rad, go use it!)
- Basic CNN model in TensorFlow (deprecated on its own branch now; I made a dogs breakfast of the git history there too, totes soz intrepid reader)
- Basically the same CNN model in Pytorch (current version)
- Basic image registration using countour detection from OpenCV2 (thanks to here for code snippets and ideas)
- Connect all the bits into a real pipeline which can be run in real time
- Experiment with reducing the number of branches in the CNN by doing colour extraction based on pixel ratios (might be better, might be worse, not sure)
- Flask plugin/API
- Mobile app