SmoothL1Loss的初步学习理解

主要参考了这一篇和这一篇博客

另外附上L1损失函数代码：

def _smooth_l1_loss(x, t, in_weight, sigma):        #sigma=1
    sigma2 = sigma ** 2
    diff = in_weight * (x - t)
    abs_diff = diff.abs()
    flag = (abs_diff.data < (1. / sigma2)).float()
    y = (flag * (sigma2 / 2.) * (diff ** 2) +
         (1 - flag) * (abs_diff - 0.5 / sigma2))
    return y.sum()