Baseline++ cross_char accuracy lower than Baseline

[yyou22]

Hello, when we run the described 5-way 5-shot experiment on Omniglot -> EMNIST in A3, the accuracy of the Baseline++ model (80%) is lower than the Baseline model's (85%). We run the code as the following: python ./train.py --dataset cross_char --num_classes 1597 for baseline python ./train.py --dataset cross_char --method baseline++ --num_classes 1597 for baseline++.

Please let us know if we need to modify any arguments or need to do anything specific for the feature saving and testing portion. Thank you so much.