Question about weight decay

[Joker316701882]

Hi, Thank you for your great work! However, there are two points in this code that I don't understand.

1. I saw 1e-7 is used as the weight decay for Adam. This is the first time of me seeing such a small weight decay for Adam. Any insights in that?
   

   lift-splat-shoot/src/train.py

   Line 42 in d74598c

   weight_decay=1e-7,

2. _I can understand using data augmentation in the training phase. Why is it still used in the testing phase? Especially cropping along the horizontal axis. Will it cause inaccurate testing results?
   

   lift-splat-shoot/src/data.py

   Lines 114 to 115 in 9d15ff2

   	crop_h = int((1 - np.mean(self.data_aug_conf['bot_pct_lim']))*newH) - fH
   	crop_w = int(max(0, newW - fW) / 2)

Looking forward to your reploy!