search

CuriousAI / mean-teacher

A state-of-the-art semi-supervised method for image recognition
https://arxiv.org/abs/1703.01780
Other
1.56k stars 331 forks source link

how to train with unlabeled data #24

Open a13668349935 opened 5 years ago

a13668349935 commented 5 years ago

i get the idea of the paper. but how can i use the unlabeled data to train the model. i see a word in the paper. if i use the unlabeled data just keep the consistency loss ,because there is no classification loss . am i right?

SHMCU commented 5 years ago

There are classification loss and consistency loss. You can set the arguments according to the command example provided in the readme.md to get the code rung. Oh, before doing that you need to run some scripts in the data-local to prepare the CIFAR10 data and labels.

tarvaina commented 5 years ago

If the question is whether mean teacher is useful without any labeled data, then answer is probably not. You should have both labeled and unlabeled data and use both classification and consistency loss.