search

gudbrandtandberg / ChessVision

Extract chess positions from photos of 2D chessboards (chess books, screenshots, etc.)
Other
33 stars 8 forks source link

FAILED_PRECONDITION finalizing GeneratorDataset #12

Open chao-mu opened 2 months ago

chao-mu commented 2 months ago

Hello,

I successfully trained the square classifier with train_square_classifier.py, but when running train_board_extractor.py, I receive the following error after training:

Epoch 64/100
60/60 [==============================] - ETA: 0s - loss: 0.0372 - dice_coeff: 0.9889 
Epoch 64: ReduceLROnPlateau reducing learning rate to 1.0000000656873453e-06.
60/60 [==============================] - 16s 275ms/step - loss: 0.0372 - dice_coeff: 0.9889 - val_loss: 0.0404 - val_dice_coeff: 0.9883 - lr: 1.0000e-05
Epoch 64: early stopping
Training the board_extractor took 17 minutes and 49 seconds
2024-07-16 18:41:24.963826: W tensorflow/core/kernels/data/generator_dataset_op.cc:107] Error occurred when finalizing GeneratorDataset iterator: FAILED_PRECONDITION: Python interpreter state is not initialized. The process may be terminated.
     [[{{node PyFunc}}]]
Error occurred when finalizing GeneratorDataset iterator: FAILED_PRECONDITION: Python interpreter state is not initialized.
gudbrandtandberg commented 1 month ago

Hey. I have never seen such an error message.. Hopefully the trained model gets written?

Anyway, I have moved the development of this project to the repo where I use pytorch instead of keras. If you are open to that, you might have more luck there.