If you run GraphPerceptionNetwork on some benckmarks, the results are as follows:

Run 10 times

the results:

{0: 0.85, 1: 0.9, 2: 0.95, 3: 0.9473684210526315, 4: 0.8947368421052632, 5: 1.0, 6: 0.9444444444444444, 7: 0.9444444444444444, 8: 1.0, 9: 0.9444444444444444}

mean:0.9375438596491229

std:0.04370581208082047

The code is still under improved and developed.

The accuracy of each folder:

{0: 0.6944444444444444, 1: 0.7222222222222222, 2: 0.7352941176470589, 3: 0.6470588235294118, 4: 0.6470588235294118, 5: 0.6764705882352942, 6: 0.6764705882352942, 7: 0.5882352941176471, 8: 0.6764705882352942, 9: 0.6176470588235294}

mean: 0.6681372549019609

std: 0.04261198567800631

The accuracy of each folder:

{0: 0.5466666666666666, 1: 0.5333333333333333, 2: 0.5533333333333333, 3: 0.5066666666666667, 4: 0.5133333333333333, 5: 0.5733333333333334, 6: 0.5733333333333334, 7: 0.5, 8: 0.56, 9: 0.54}

mean: 0.5400000000000001

std: 0.025121924908555707

The accuracy of each folder:

{0: 0.7589285714285714, 1: 0.8214285714285714, 2: 0.8125, 3: 0.8018018018018018, 4: 0.8108108108108109, 5: 0.7747747747747747, 6: 0.7477477477477478, 7: 0.7477477477477478, 8: 0.7837837837837838, 9: 0.8198198198198198}

mean:0.787934362934363

std:0.027783661934815667

The results may vary slightly.