search

dragen1860 / MAML-Pytorch

Elegant PyTorch implementation of paper Model-Agnostic Meta-Learning (MAML)
MIT License
2.31k stars 420 forks source link

About accuracy #73

Open feisichen opened 1 year ago

feisichen commented 1 year ago

ec2e2e5eb9e69b0cd678e1af2f1ffa33

I trained the model using my own dataset.Does this mean that I have achieved 79.6% accuracy in few-shot classification? But I have only trained for ten minutes. . .I'm sure I didn't put the test set into the training set.

sevenHsu commented 1 year ago

Actually，your test set will be finetuned (train 10 steps for each support set) after every 500 steps training on training set. So you can see the acc growth from 0,2615 to 0.796 via 10 steps finetune.

feisichen commented 1 year ago

Thank you. Are the support and query sets automatically split from the test set?

sevenHsu commented 1 year ago

They were randomly selected from each task test dataset.

LittleShuo commented 1 year ago

Hello, where did you get those csv files in your miniimagenet dataset?

feisichen commented 1 year ago

Hello, where did you get those csv files in your miniimagenet dataset?

Sorry, I didn't use the miniimagenet dataset, so I'm not very familiar with it. I'm using my own dataset, where the images for each category are placed in their respective folders. So, I wrote a simple Python program to generate the csv files.

LittleShuo commented 1 year ago

Ok, thank you very much for your reply.