search

CuriousAI / mean-teacher

A state-of-the-art semi-supervised method for image recognition
https://arxiv.org/abs/1703.01780
Other
1.58k stars 331 forks source link

The accuracy of validation data is very low #52

Open zhang-qiang-github opened 3 years ago

zhang-qiang-github commented 3 years ago

Thank you for sharing this project.

I run the cifar10_test.py with the default configuration, and the accuracy of training data is about 50% for top 1and 90% for top 5.

However, for validation data, the accuracy is very low, and the accuracy is 0.154% for top1/5. Please have a look for the following figure.

image

Any suggestion is appreciated!