search

ybendou / easy

This repository is the official implementation Ensemble Augmented-Shot Y-shaped Learning: State-Of-The-Art Few-Shot Classification with Simple Ingredients.
MIT License
112 stars 18 forks source link

Other backbones #5

Closed Qianshaowei closed 2 years ago

Qianshaowei commented 2 years ago

Thanks for your work. Do you have the weight files of other backbone networks that you have trained? Such as WideResNet28 and ResNet10.

ybendou commented 2 years ago

Hello,

We only train for ResNet12 in the paper. However, the code is ready for WideResNet28 and you can use it to train a WRN-28 by setting --model 'wideresnet' in the training command.

Hope this helps,

Best,

Qianshaowei commented 2 years ago

Ok, I know this command, but it takes too much time to train a pretrained model. Thanks for your reply.

shepherdls commented 1 year ago

hello, i have a question. for the wideresnet28 and fc100, the following command can be ok? python main.py --dataset-path /home/dataset --dataset FC100 --model wideresnet --milestones 100 --epochs 0 --manifold-mixup 600 --cosine --gamma 0.9 --rotation --batch-size 64 --device cuda: 0 --preprocessing ME --skip-epochs 550 --save-model /home/1.pt1 --n-shots 5 what's the setting of --epochs, manifold-mixup and --skip-epochs ,the values are right?