google-research / augmix

AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
Apache License 2.0
979 stars 157 forks source link

Can you share any acc information per epoch? #7

Closed seominseok0429 closed 4 years ago

seominseok0429 commented 4 years ago

Hi Author I am a college student studying deep learning in Korea. I read your paper very interesting. But I am running AugMix on imagenet and I have to wait 6 days with my GPU machine. So I'm wondering if learning is going well. To find out the trend in the middle, can you share any acc information per epoch?

hendrycks commented 4 years ago

Here's the training log of a model that trained for 90 epochs. https://drive.google.com/file/d/1CGHCaK-ol6UjVsBFo-87m18WA6sq25qJ/view?usp=sharing

seominseok0429 commented 4 years ago

Thank you for your kind reply. Do you ever get a lot of performance instability in the early stages of learning?

hendrycks commented 4 years ago

No. Do you have better luck with this code? https://github.com/google-research/augmix/issues/3#issuecomment-569455385 If so, that that suggests the main code might have an issue, but the code I linked to I have certainly used without instability.