In the code, you report the best accuracy on the test set over all epochs. But generally, the validation set should be used for model selection and not the test set. I am aware that the validation set is not available for CUB-200-2011. Is this the standard practice for this dataset? There are other works that use a subset of the training set for validation. In that case, the comparison may not be fair.
In the code, you report the best accuracy on the test set over all epochs. But generally, the validation set should be used for model selection and not the test set. I am aware that the validation set is not available for CUB-200-2011. Is this the standard practice for this dataset? There are other works that use a subset of the training set for validation. In that case, the comparison may not be fair.