search

yaoyao-liu / meta-transfer-learning

TensorFlow and PyTorch implementation of "Meta-Transfer Learning for Few-Shot Learning" (CVPR2019)
https://lyy.mpi-inf.mpg.de/mtl/
MIT License
731 stars 147 forks source link

“MAX_ITER=20000” during meta-test in Tensorflow code #58

Open licj1 opened 2 years ago

licj1 commented 2 years ago

Hello, Thanks for your excellent work.

I have a question about "MAX_ITER = 20000" during meta-test in TensorFlow, MAML's meta-test was only done once, why do you need to do yours for MAX_ITER times? And in this way, how to get the final results? image

Thanks for your time and look forward to your reply soon!

yaoyao-liu commented 2 years ago

Thanks for your interest in our work!

Running the following loop won't execute the meta-test for MAX_ITER times. It will only execute the meta-test for each saved model: https://github.com/yaoyao-liu/meta-transfer-learning/blob/5c12cc48042a35198b114de49f24199722e74cbe/tensorflow/run_experiment.py#L105-L106

If you have any further comments, please feel free to add comments here.

licj1 commented 2 years ago

Thanks for your interest in our work!

Running the following loop won't execute the meta-test for MAX_ITER times. It will only execute the meta-test for each saved model:

https://github.com/yaoyao-liu/meta-transfer-learning/blob/5c12cc48042a35198b114de49f24199722e74cbe/tensorflow/run_experiment.py#L105-L106

If you have any further comments, please feel free to add comments here.

Thanks for your prompt reply. So can I perform the meta-test based on the saved model with the last iteration and use this result as the final result?

yaoyao-liu commented 2 years ago

@licj1

You may see the meta-test accuracy for each saved model, respectively. I'm not sure what you mean by "final result". If you hope to know which models correspond to the results reported in the paper, you may see "Section 5.3 Results and analysis".

Please feel free to add comments here if you have any further comments.