Your experimental results are only accidental and not convincing

[giao-cloude]

May I ask if you could disclose the composition of your training and validation sets? I suspect that you only used a clever method to achieve high accuracy, and changing the validation set would not achieve your experimental results