Chess piece image recognizer.
In this repository, a chess piece image recognizer is created using a multilayer neural network trained on the 2D Chessboard and Chess Pieces dataset.
For further information, please visit:
Perception is the building block that allows to recognize the piece placement in FEN format of a chessboard image as it is implemented in the Chess\Media namespace of the PHP Chess library.
Example:
Clone the chesslablab/perception repo into your projects folder:
git clone [email protected]:chesslablab/perception.git
Then cd the perception directory and install the Composer dependencies:
composer install
Make sure to remove all files in the testing and the training folders.
php cleanup.php
Prepare 50,000 samples for further training.
php prepare.php 50000 training
Train the neural network.
php train.php
Prepare 20,000 samples for further testing.
php prepare.php 20000 testing
Make predictions.
php validate.php
Below is an excerpt from an example report.
{
"breakdown": {
"overall": {
"accuracy": 0.9994510579306761,
"balanced accuracy": 0.9977038971505224,
"f1 score": 0.9961001544839454,
"precision": 0.9965510770350279,
"recall": 0.9957062828647335,
"specificity": 0.9997015114363115,
"negative predictive value": 0.9997024767903292,
"false discovery rate": 0.003448922964972234,
"miss rate": 0.004293717135266537,
"fall out": 0.00029848856368840636,
"false omission rate": 0.0002975232096707342,
"mcc": 0.9958186303218224,
"informedness": 0.9954077943010451,
"markedness": 0.9962535538253571,
"true positives": 49819,
"true negatives": 597828,
"false positives": 178,
"false negatives": 178,
"cardinality": 49997
},
"classes": {
"1": {
"accuracy": 0.9996990006822651,
"balanced accuracy": 0.9921293527894939,
"f1 score": 0.9908814589665653,
"precision": 0.9975520195838433,
"recall": 0.9842995169082126,
"specificity": 0.999959188670775,
"negative predictive value": 0.9997347858906094,
"false discovery rate": 0.002447980416156681,
"miss rate": 0.0157004830917874,
"fall out": 4.081132922495456e-5,
"false omission rate": 0.0002652141093906213,
"informedness": 0.9842587055789878,
"markedness": 0.9972868054744528,
"mcc": 0.9907513412795808,
"true positives": 815,
"true negatives": 49004,
"false positives": 2,
"false negatives": 13,
"cardinality": 828,
"proportion": 0.016560993659619577
},
"B": {
"accuracy": 0.9984968132440775,
"balanced accuracy": 0.9989570822259015,
"f1 score": 0.9908301748379998,
"precision": 0.9823030303030303,
"recall": 0.9995066600888012,
"specificity": 0.9984075043630017,
"negative predictive value": 0.9999563023006839,
"false discovery rate": 0.017696969696969655,
"miss rate": 0.0004933399111988201,
"fall out": 0.001592495636998259,
"false omission rate": 4.3697699316114225e-5,
"informedness": 0.997914164451803,
"markedness": 0.9822593326037143,
"mcc": 0.9900558070988832,
"true positives": 4052,
"true negatives": 45767,
"false positives": 73,
"false negatives": 2,
"cardinality": 4054,
"proportion": 0.08108486509190552
},
"Q": {
"accuracy": 0.998016747465844,
"balanced accuracy": 0.9989202041795733,
"f1 score": 0.9880014543691674,
"precision": 0.9762874251497006,
"recall": 1,
"specificity": 0.9978404083591467,
"negative predictive value": 1,
"false discovery rate": 0.023712574850299428,
"miss rate": 0,
"fall out": 0.002159591640853331,
"false omission rate": 0,
"informedness": 0.9978404083591466,
"markedness": 0.9762874251497005,
"mcc": 0.9870050876197534,
"true positives": 4076,
"true negatives": 45743,
"false positives": 99,
"false negatives": 0,
"cardinality": 4076,
"proportion": 0.08152489149348961
},
"b": {
"accuracy": 1,
"balanced accuracy": 1,
"f1 score": 1,
"precision": 1,
"recall": 1,
"specificity": 1,
"negative predictive value": 1,
"false discovery rate": 0,
"miss rate": 0,
"fall out": 0,
"false omission rate": 0,
"informedness": 1,
"markedness": 1,
"mcc": 1,
"true positives": 4078,
"true negatives": 45741,
"false positives": 0,
"false negatives": 0,
"cardinality": 4078,
"proportion": 0.08156489389363362
},
"K": {
"accuracy": 0.9988771929824561,
"balanced accuracy": 0.9936700067149988,
"f1 score": 0.9931906614785991,
"precision": 0.9990215264187867,
"recall": 0.9874274661508704,
"specificity": 0.9999125472791273,
"negative predictive value": 0.998864306462533,
"false discovery rate": 0.0009784735812132794,
"miss rate": 0.012572533849129597,
"fall out": 8.745272087273825e-5,
"false omission rate": 0.0011356935374670085,
"informedness": 0.9873400134299977,
"markedness": 0.9978858328813196,
"mcc": 0.9925989178105358,
"true positives": 4084,
"true negatives": 45735,
"false positives": 4,
"false negatives": 52,
"cardinality": 4136,
"proportion": 0.08272496349780988
},
"k": {
"accuracy": 1,
"balanced accuracy": 1,
"f1 score": 1,
"precision": 1,
"recall": 1,
"specificity": 1,
"negative predictive value": 1,
"false discovery rate": 0,
"miss rate": 0,
"fall out": 0,
"false omission rate": 0,
"informedness": 1,
"markedness": 1,
"mcc": 1,
"true positives": 4154,
"true negatives": 45665,
"false positives": 0,
"false negatives": 0,
"cardinality": 4154,
"proportion": 0.08308498509910595
},
"N": {
"accuracy": 0.9993781344032097,
"balanced accuracy": 0.9961700024709661,
"f1 score": 0.9961552771921122,
"precision": 1,
"recall": 0.9923400049419323,
"specificity": 1,
"negative predictive value": 0.9993236462015098,
"false discovery rate": 0,
"miss rate": 0.007659995058067692,
"fall out": 0,
"false omission rate": 0.0006763537984901813,
"informedness": 0.9923400049419322,
"markedness": 0.9993236462015098,
"mcc": 0.9958257036300057,
"true positives": 4016,
"true negatives": 45803,
"false positives": 0,
"false negatives": 31,
"cardinality": 4047,
"proportion": 0.08094485669140149
},
"n": {
"accuracy": 1,
"balanced accuracy": 1,
"f1 score": 1,
"precision": 1,
"recall": 1,
"specificity": 1,
"negative predictive value": 1,
"false discovery rate": 0,
"miss rate": 0,
"fall out": 0,
"false omission rate": 0,
"informedness": 1,
"markedness": 1,
"mcc": 1,
"true positives": 4242,
"true negatives": 45577,
"false positives": 0,
"false negatives": 0,
"cardinality": 4242,
"proportion": 0.08484509070544233
},
"P": {
"accuracy": 0.9987570417593873,
"balanced accuracy": 0.9925444925444926,
"f1 score": 0.9924884904288829,
"precision": 1,
"recall": 0.9850889850889851,
"specificity": 1,
"negative predictive value": 0.9986458447089658,
"false discovery rate": 0,
"miss rate": 0.014911014911014875,
"fall out": 0,
"false omission rate": 0.0013541552910342203,
"informedness": 0.9850889850889852,
"markedness": 0.9986458447089657,
"mcc": 0.991844252706889,
"true positives": 4096,
"true negatives": 45723,
"false positives": 0,
"false negatives": 62,
"cardinality": 4158,
"proportion": 0.08316498989939396
},
"p": {
"accuracy": 1,
"balanced accuracy": 1,
"f1 score": 1,
"precision": 1,
"recall": 1,
"specificity": 1,
"negative predictive value": 1,
"false discovery rate": 0,
"miss rate": 0,
"fall out": 0,
"false omission rate": 0,
"informedness": 1,
"markedness": 1,
"mcc": 1,
"true positives": 4167,
"true negatives": 45652,
"false positives": 0,
"false negatives": 0,
"cardinality": 4167,
"proportion": 0.083345000700042
},
"q": {
"accuracy": 1,
"balanced accuracy": 1,
"f1 score": 1,
"precision": 1,
"recall": 1,
"specificity": 1,
"negative predictive value": 1,
"false discovery rate": 0,
"miss rate": 0,
"fall out": 0,
"false omission rate": 0,
"informedness": 1,
"markedness": 1,
"mcc": 1,
"true positives": 4000,
"true negatives": 45819,
"false positives": 0,
"false negatives": 0,
"cardinality": 4000,
"proportion": 0.08000480028801728
},
"R": {
"accuracy": 0.9996388225615507,
"balanced accuracy": 0.9977595220313666,
"f1 score": 0.9977544910179641,
"precision": 1,
"recall": 0.9955190440627334,
"specificity": 1,
"negative predictive value": 0.9996073127099786,
"false discovery rate": 0,
"miss rate": 0.004480955937266606,
"fall out": 0,
"false omission rate": 0.00039268729002139846,
"informedness": 0.9955190440627333,
"markedness": 0.9996073127099785,
"mcc": 0.9975610840380431,
"true positives": 3999,
"true negatives": 45820,
"false positives": 0,
"false negatives": 18,
"cardinality": 4017,
"proportion": 0.08034482068924136
},
"r": {
"accuracy": 1,
"balanced accuracy": 1,
"f1 score": 1,
"precision": 1,
"recall": 1,
"specificity": 1,
"negative predictive value": 1,
"false discovery rate": 0,
"miss rate": 0,
"fall out": 0,
"false omission rate": 0,
"informedness": 1,
"markedness": 1,
"mcc": 1,
"true positives": 4040,
"true negatives": 45779,
"false positives": 0,
"false negatives": 0,
"cardinality": 4040,
"proportion": 0.08080484829089746
}
}
},
"matrix": {
"1": {
"1": 815,
"B": 0,
"Q": 0,
"b": 0,
"K": 0,
"k": 0,
"N": 0,
"n": 0,
"P": 2,
"p": 0,
"q": 0,
"R": 0,
"r": 0
},
"B": {
"1": 13,
"B": 4052,
"Q": 0,
"b": 0,
"K": 0,
"k": 0,
"N": 0,
"n": 0,
"P": 60,
"p": 0,
"q": 0,
"R": 0,
"r": 0
},
"Q": {
"1": 0,
"B": 2,
"Q": 4076,
"b": 0,
"K": 52,
"k": 0,
"N": 27,
"n": 0,
"P": 0,
"p": 0,
"q": 0,
"R": 18,
"r": 0
},
"b": {
"1": 0,
"B": 0,
"Q": 0,
"b": 4078,
"K": 0,
"k": 0,
"N": 0,
"n": 0,
"P": 0,
"p": 0,
"q": 0,
"R": 0,
"r": 0
},
"K": {
"1": 0,
"B": 0,
"Q": 0,
"b": 0,
"K": 4084,
"k": 0,
"N": 4,
"n": 0,
"P": 0,
"p": 0,
"q": 0,
"R": 0,
"r": 0
},
"k": {
"1": 0,
"B": 0,
"Q": 0,
"b": 0,
"K": 0,
"k": 4154,
"N": 0,
"n": 0,
"P": 0,
"p": 0,
"q": 0,
"R": 0,
"r": 0
},
"N": {
"1": 0,
"B": 0,
"Q": 0,
"b": 0,
"K": 0,
"k": 0,
"N": 4016,
"n": 0,
"P": 0,
"p": 0,
"q": 0,
"R": 0,
"r": 0
},
"n": {
"1": 0,
"B": 0,
"Q": 0,
"b": 0,
"K": 0,
"k": 0,
"N": 0,
"n": 4242,
"P": 0,
"p": 0,
"q": 0,
"R": 0,
"r": 0
},
"P": {
"1": 0,
"B": 0,
"Q": 0,
"b": 0,
"K": 0,
"k": 0,
"N": 0,
"n": 0,
"P": 4096,
"p": 0,
"q": 0,
"R": 0,
"r": 0
},
"p": {
"1": 0,
"B": 0,
"Q": 0,
"b": 0,
"K": 0,
"k": 0,
"N": 0,
"n": 0,
"P": 0,
"p": 4167,
"q": 0,
"R": 0,
"r": 0
},
"q": {
"1": 0,
"B": 0,
"Q": 0,
"b": 0,
"K": 0,
"k": 0,
"N": 0,
"n": 0,
"P": 0,
"p": 0,
"q": 4000,
"R": 0,
"r": 0
},
"R": {
"1": 0,
"B": 0,
"Q": 0,
"b": 0,
"K": 0,
"k": 0,
"N": 0,
"n": 0,
"P": 0,
"p": 0,
"q": 0,
"R": 3999,
"r": 0
},
"r": {
"1": 0,
"B": 0,
"Q": 0,
"b": 0,
"K": 0,
"k": 0,
"N": 0,
"n": 0,
"P": 0,
"p": 0,
"q": 0,
"R": 0,
"r": 4040
}
}
}
We encourage you to contribute to the Perception repository! Please follow the Contributing Guidelines.
Made with contrib.rocks
The MIT license.