robrank
robrank copied to clipboard
Name 'os' is not defined.
Hi, I'm trying to do a defensive training on resnet18 without SPC2 term so that I can futher alter the loss function.
However, I'm constantly encountering with the following problem, but I have no clue about where could possbily be wrong.

Thanks!
Thank you for reporting this issue. To build this work upon the ICML work, SPC-2 batch is used by default. The ttripletN config is inherited from the old implementation, and hence may suffer from some unfixed bugs. I may look into it in the future and fix the problem, and any patch will be appreciated as well.
BTW, as mentioned in the ICML20 paper "revisiting ...", SPC-2 batch reaches better result than simple random batches. So do we have a reason to seriously support the old function?