Unexpected CIFAR10 Results

[maordikter1]

Hi!
First, thanks for your great work!
I followed your instructions and tried running labo_train on CIFAR10. However, I'm running into an issue with the results.
I ran labo_train on the dataset and received the following results:
1 shot: 86.34% on val, 85.39% on test.
2 shot: 90.64% on val, 89.68% on test.
4 shot: 91.04% on val, 91.10% on test.
8 shot: 93.44% on val, 92.9% on test.
16 shot: 94.84% on val, 94.80% on test.

However these results do not match the results in the paper, especially in the fewer shots.
Do you have any idea why this might be happening?

Thanks

[YueYANG1996]

How many epochs did you train? CIFAR-10 needs to be trained longer, and the validation accuracy continues to increase slowly (see figure below). Also, could you let me know if you changed any hyperparameters in the config file?

[maordikter1]

Thanks for the quick response!
I think I solved the problem. The problem was in the additional code for preprocessing CIFAR10.
I deleted the last 2 lines that saves the splits, It seems that these lines were overwriting the provided splits and causing modifications.

Thanks