About the final XBNBlock-P2 module

[yangyahu-1994]

Dear Professor Huang:
Hello! Is the final XBNBlock-P2 module the following code?

def GroupNorm(num_features, num_groups=32, eps=1e-5, affine=True, *args, **kwargs):
    if num_groups>num_features:
        print('------arrive maxum groub numbers of:', num_features)
        num_groups=num_features
    return nn.GroupNorm(num_groups, num_features, eps=eps, affine=affine)


class Bottleneck_XBNBlock_P2(nn.Module):
        expansion = 4
    
        def __init__(self, inplanes, planes, stride=1, downsample=None):
            super(Bottleneck_XBNBlock_P2, self).__init__()
            self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
            self.bn1 = nn.BatchNorm2d(planes)
            self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
            self.bn2 = GroupNorm(planes)
            self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
            self.bn3 = nn.BatchNorm2d(planes * 4)
            self.relu = nn.ReLU(inplace=True)
            self.downsample = downsample
            self.stride = stride

        def forward(self, x):
            residual = x
    
            out = self.conv1(x)
            out = self.bn1(out)
            out = self.relu(out)
    
            out = self.conv2(out)
            out = self.bn2(out)
            out = self.relu(out)
    
            out = self.conv3(out)
            out = self.bn3(out)
    
            if self.downsample is not None:
                residual = self.downsample(x)
    
            out += residual
            out = self.relu(out)
    
            return out

I read your code and only this class matches the module above.
Also, I use the standard Bottleneck in my work: conv 11->bn->relu->conv33->bn->relu. The structure of this Bottleneck is different from the one in your article. Can I replace the bn after 3*3 convolution with BFN(GN)?
Thank you very much.

[huangleiBuaa]

Hi, yes, the module you list is the XBNBlock-P2. You can replace the BN (after 33 convolution) with BFN in the standard 'basicBlock' (based on your description, you have only 2 conv, I am not sure whether you forget the other 11 conv), technicually. However, based on my experience on the Bottleneck, it seems that repalce the BN (after 1*1 convolution) with BFN should be better in your standard 'basicBlock'.

[yangyahu-1994]

非常感谢黄老师的耐心解答。
我的Bottleneck如下图：

确实只有两个Conv。我前几天将33convolution之后的BN换成了BFN，效果确实没有以前好了。我再试试11convolution之后的BN换成BFN。
谢谢黄老师。

[Sakura0v01]

同学你好，我想请问一下关于下载下来的代码ResNext_XBNBlock_P2.py中model_urls没有定义是如何解决的