Patio Corners
A lightweight PyTorch CNN for recognizing visible playing-card corners in Poker Patio screenshots. It classifies 52 rank-and-suit identities plus an unknown class, using grayscale crops resized to 32 × 48 pixels.
patio-corners.pt contains the trained CornerNet state dictionary and class labels. It was trained on augmented Poker Patio card assets and negative UI/card-back examples for a local poker OCR portfolio project. Use it with the matching CornerNet architecture and card-region preprocessing; it is not a standalone full-image detector.
For educational purposes only. The model recognizes cards, not poker strategy. Accuracy on other websites, themes or unseen capture conditions is not established.