densechen

``` class NeuNorm(nn.Module): def __init__(self, in_channels, height, width, k=0.9): super().__init__() self.x = 0 self.k0 = k self.k1 = (1 - self.k0) / in_channels**2 # self.w = nn.Parameter(torch.Tensor(in_channels, height, width)) self.w...

in_channels or 1 will make it not share or share the weight among different channels. Both of them may be ok. 🤔 [in_channels, height, width] may be more consistend with...

The author have released a code [here](https://github.com/yjwu17/BP-for-SpikingNN). However, they did not provide the implementation of NeuNorm. [issue](https://github.com/yjwu17/BP-for-SpikingNN/issues/1)

This is really confusing me a lot. Did you try to fold the batch norm layer? What the performance is?

> I got the mAP: 0.5145 me, too. What's wrong on earth?

Refer to here: [issue](https://github.com/eriklindernoren/PyTorch-YOLOv3/issues/119)

I also think so!

我想问一下，你们训练出来的精度，和作者readme里面给的精度比起来吻合吗？ 我训练出来的精度普遍高了1-1.5个百分点

@ww-zwj 谢谢！

I also notice that. I think the main difference is that resume from checkpoint will cotinuous training from the best state of model, but scheduler will continuous training with current...