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

Argument setting for training other datasets #18

Closed LIUZIJING-CHN closed 1 year ago

LIUZIJING-CHN commented 2 years ago

Sorry to bother again. I wonder if you can provide the argument setting for reproducing outcomes of the other datasets like CIFAR-FS or Tiered-ImageNet. Hope this won't bring you much trouble. Thx a lot.

ybendou commented 2 years ago

Hello,

For tiered you can use the following command:

python main.py --dataset tieredimagenet --model resnet12 --milestones 300 --epochs 0 --manifold-mixup 1500 --cosine --gamma 0.9 --rotations --batch-size 128 --device cuda:0 --preprocessing ""ME"" --dataset-size 12800 --skip-epochs 1450 --deterministic

For cifarfs:

python main.py --dataset cifarfs --model resnet12 --milestones 300 --epochs 0 --manifold-mixup 1500 --cosine --gamma 0.9 --rotations --batch-size 128 --device cuda:0 --preprocessing ""ME"" --dataset-size 12800 --skip-epochs 1450 --deterministic

You also need to add the path of your datasets folder --dataset-path

Let me know if i doesn't work.

Brest, Yassir

LIUZIJING-CHN commented 2 years ago

Thx for such a quick reply! I'll try them soon.

LIUZIJING-CHN commented 2 years ago

By the way, the argument for CIFAR-FS may be wrong. I wonder if you directly copy that for tiered-ImageNet?

ybendou commented 2 years ago

Sorry for the typo, I modified it. Indeed, it's the same parameters for both datasets, the only difference is --dataset cifarfs instead.

LIUZIJING-CHN commented 1 year ago

sorry to bother again, but what about CUB and FC-100? I am not meant to bother you, my project depends on your code, and I need a full reproduction. Sorry again

ybendou commented 1 year ago

Hello,

No worries, regarding the data choices, it's using the --dataset parameter and these are all the possible options: ['miniimagenet', 'cubfs','tieredimagenet', 'fc100', 'cifarfs']. So for CUB it's cubfs and FC-100 it's fc100.

Hope this helps,

Best,

LIUZIJING-CHN commented 1 year ago

I mean the important arguments like dataset-size & manifold epoch, are they the same as that of mini-imagenet or tieredimagenet?

ybendou commented 1 year ago

Hello,

Sorry for the late answer, I double checked and the right parameters for cubfs, fc100 and cifarfs100 datasets should be the following: --manifold-mixup 600 --milestones 100 --skip-epochs 550 without the --dataset-size option:

Only miniimagenet and tieredimagenet are trained for 1500 epochs with the dataset-size option.

LIUZIJING-CHN commented 1 year ago

I see，THX again！