Chess Piece Classifier (Custom CNN)
A from-scratch CNN that classifies 50x50 chess-board cells into 13 classes
(empty + 6 white + 6 black pieces), trained on the synthetic
koryakinp/chess-positions dataset with Kornia GPU augmentation and Optuna
hyperparameter tuning.
Classes
- 0: empty
- 1: wP
- 2: wN
- 3: wB
- 4: wR
- 5: wQ
- 6: wK
- 7: bP
- 8: bN
- 9: bB
- 10: bR
- 11: bQ
- 12: bK
Usage
import torch
from huggingface_hub import hf_hub_download
weights = hf_hub_download(repo_id="honi05/chess-piece-cnn", filename="chess_cnn.pt")
state = torch.load(weights, map_location="cpu")