search

MedMNIST / experiments

Codebase for reproducible benchmarking experiments in MedMNIST v2
Apache License 2.0
44 stars 18 forks source link

Doubt #1

Closed agaldran closed 2 years ago

agaldran commented 2 years ago

Hi!

Thanks so much for making this resource available, it is very useful as a quick benchmark for new ideas. I was wondering if you used in your papers (v1 and v2) resnets trained started from Imagenet weights? It seems from your code that the default is False, so they would be trained from scratch, but the papers do not mention this detail, I think, and I just wanted to double-check with you..

Thanks!

duducheng commented 2 years ago

Hi @agaldran,

We don’t use pretraining to avoid any confounders in analysis. We have also tried the pretraining, there seems no significant difference (but we don’t have statistic analysis). It should be very easy to try if you need to know ;)

cc: @ThreeSRR

Cheers, Jiancheng

agaldran commented 2 years ago

That's alright, thanks Jiancheng for the quick reply and the info, and thanks again for the resource.