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
736 stars 149 forks source link

Very low performance during pre-training phase on FC100 #54

Open TristanGomez44 opened 3 years ago

TristanGomez44 commented 3 years ago

Hi Yaoyao, Thank you for this very interesting work and code.

However, I have an issue with the pre-training phase on FC100, the validation performance is very low: 33%, even after several runs to check with various seeds.

Here is the command I run : python3 main.py --dataset_dir datasets/FC100/ --phase pre_train --gpu 0

And here is the results I get : low_perf

Is the performance supposed to be this low ?

I use pytorch 1.8.1 and torchvision 0.9.1

Kind regards,

Tristan.

yaoyao-liu commented 3 years ago

Thanks for your interest in our work. I will check the pre-training code for FC100 later.

You may try to use the pre-training code in this project. I have tested it, and I think it would work.

Best, Yaoyao