Potential bug in 2 image augmentation code for base step training

[prachigarg23]

Hi @CanPeng123

In train.py > train_two_image() : Line 159 angular_loss = 0.5 * angular_loss_1 + 0.5 + angular_loss_2

Is the + 0.5 a bug? In the paper it says that loss for 2 augmentation images is averaged before back propagation.

[CanPeng123]

Hi, Yes, it is a typo mistake when I cleaned and uploaded the code. I have changed it. Sorry for the inconvenience.