bug about losslandscapematters HOT 1 OPEN

Hi @13luoyu,

We do not recommend use this code or (https://github.com/huanzhang12/CROWN-IBP/blob/master/train.py) to run CAP (https://arxiv.org/abs/1711.00851). We recommend using the original implementation (https://github.com/locuslab/convex_adversarial) for CAP.

Quick Fix:
You may comment out the lines 203 and 441 where DualNetwork is used, or delete the whole CAP parts

if kwargs["bound_type"] == "convex-adv":
    ...

We will fix the bug accordingly.

Thanks,
Sungyoon Lee

from losslandscapematters.