search

nupurkmr9 / S2M2_fewshot

Other
112 stars 20 forks source link

How to reproduce your result? #13

Closed ljjcoder closed 3 years ago

ljjcoder commented 3 years ago

Thanks for your work! I want to reproduce the results of Rolation. I followed your request for three steps. First： python train.py --dataset miniImagenet --method rotation --model WideResNet28_10 --stop_epoch 200 --batch_size 64 --test_batch_size 16 second： python save_features.py --dataset miniImagenet --method rotation --model WideResNet28_10 Third： python test.py --dataset miniImagenet --method rotation --model WideResNet28_10 --n_shot 1 However, I only got an accuracy of 0.56, while your paper has an accuracy of 0.639. Are there key steps missing in my training process?

nupurkmr9 commented 3 years ago

Hi, Sorry for not mentioning in the repo, but the miniImagenet model was trained for 400 epochs. Thanks for pointing this out, I will update the repo with scripts for replicating results for all datasets. Let us know if you are able to replicate the results for rotation now.