About the data augmentation

[skyhat3]

Thank you for your outstanding work. During my reproduction of your experiments, I have been unable to achieve the reported high performance metrics by run scripts\maple\base2new_train_maple.sh
Instead, I constructed a custom pipeline independently and obtained results close to those presented in your paper.
I noticed that the scale parameter of random_resized_crop was not recorded in the training logs you provided, while in the latest version of your code, this parameter is set to (0.08, 1) during training. And now I set it to (0.5,1).
yours

mine

I would greatly appreciate it if you could kindly confirm whether there was a modification to this parameter, as this detail may have a significant impact on the experimental results.
I would highly appreciate it if you could kindly provide more details regarding the training configuration.