DGCN icon indicating copy to clipboard operation
DGCN copied to clipboard

你的normalization()函数好像并没有起到normal的作用

Open C3Shirt opened this issue 3 years ago • 1 comments

mean = train.mean(axis=0, keepdims=True)
std = train.std(axis=0, keepdims=True)

def normalize(x):
    return (x - mean) / std

train = (train).transpose(0,2,1,3)
val = (val).transpose(0,2,1,3)
test =(test).transpose(0,2,1,3)

return {'mean': mean, 'std': std}, train, val, test

虽然这里写了 normalize(x),但是并没有作用到训练集上

C3Shirt avatar Sep 23 '21 08:09 C3Shirt

mean = train.mean(axis=0, keepdims=True)
std = train.std(axis=0, keepdims=True)

def normalize(x):
    return (x - mean) / std

train = (train).transpose(0,2,1,3)
val = (val).transpose(0,2,1,3)
test =(test).transpose(0,2,1,3)

return {'mean': mean, 'std': std}, train, val, test

虽然这里写了 normalize(x),但是并没有作用到训练集上

对,确实没在这里归一化,在model.py里的DGCN block的输入上直接使用了 x_w=self.bn(x_w) x_d=self.bn(x_d) x_r=self.bn(x_r) 这里相当于实时归一化了

guokan987 avatar Sep 23 '21 16:09 guokan987