search

zhangdoudou123 / SemFew

[CVPR2024] Simple Semantic-Aided Few-Shot Learning
16 stars 2 forks source link

Related hyperparameters for FC100 training #2

Closed pengshy closed 1 week ago

pengshy commented 3 weeks ago

Thank you for your code. This is an outstanding job. But I cannot reproduce the results of paper FC100 based on the parameters of training two ImageNet datasets. Could you please provide me with the parameter settings for training FC100.

zhangdoudou123 commented 3 weeks ago

The experimental results are related to the training of the backbone network, with no other particular hyper-parameters.

pengshy commented 2 weeks ago

Thanks for your reply. I follow your steps but the accuracy of FC100 is still about 45%. How should I solve it？

zhangdoudou123 commented 2 weeks ago

You could change the backbone, for example, resnet or transformer.

DruryXu commented 2 weeks ago

Please try to reproduce the results based on weights we uploaded today, sorry for any inconvenience caused and feel free to contact us if there are any new issues.

pengshy commented 2 weeks ago

Thank you so much!!!!