The code does not match the implementation details in the paper

med-air / DLTTA

[IEEE TMI'22] DLTTA: Dynamic Learning Rate for Test-time Adaptation on Cross-domain Medical Images

16 stars 2 forks source link

The code does not match the implementation details in the paper #8

Open wwh13843218306 opened 3 months ago

wwh13843218306 commented 3 months ago

The paper mentions the feature channel dimensions are [16,32,64,128], in reality on the OCT dataset your code is [64, 64, 64, 64]。And you didn't mention the part about ANET and AENET in the paper.

HongzhengYang commented 3 months ago

AENET is for ATTA method. For the feature channels, do you mean the feature dimension for the UNET block ?

wwh13843218306 commented 3 months ago

AENET用于ATTA方法。对于特征通道，您的意思是 UNET 块的特征维度吗？

Yes, I found that unet's feature dimensions don't match the paper。 Did I misunderstand something?

HongzhengYang commented 3 months ago

Hi, I think you can modify the unet_dims argument to match the feature channels dimension. Or you can directly use the default configuration.