Any difference between Attention and SAGAN?

[jayzhan211]

https://github.com/heykeetae/Self-Attention-GAN/blob/master/sagan_models.py In this code

class Self_Attn(nn.Module):
    """ Self attention Layer"""
    def __init__(self,in_dim,activation):
        super(Self_Attn,self).__init__()
        self.chanel_in = in_dim
        self.activation = activation

        self.query_conv = nn.Conv2d(in_channels = in_dim , out_channels = in_dim//8 , kernel_size= 1)
        self.key_conv = nn.Conv2d(in_channels = in_dim , out_channels = in_dim//8 , kernel_size= 1)
        self.value_conv = nn.Conv2d(in_channels = in_dim , out_channels = in_dim , kernel_size= 1)
        self.gamma = nn.Parameter(torch.zeros(1))

        self.softmax  = nn.Softmax(dim=-1) #
    def forward(self,x):
        """
            inputs :
                x : input feature maps( B X C X W X H)
            returns :
                out : self attention value + input feature 
                attention: B X N X N (N is Width*Height)
        """
        m_batchsize,C,width ,height = x.size()
        proj_query  = self.query_conv(x).view(m_batchsize,-1,width*height).permute(0,2,1) # B X CX(N)
        proj_key =  self.key_conv(x).view(m_batchsize,-1,width*height) # B X C x (*W*H)
        energy =  torch.bmm(proj_query,proj_key) # transpose check
        attention = self.softmax(energy) # BX (N) X (N) 
        proj_value = self.value_conv(x).view(m_batchsize,-1,width*height) # B X C X N

        out = torch.bmm(proj_value,attention.permute(0,2,1) )
        out = out.view(m_batchsize,C,width,height)

        out = self.gamma*out + x
        return out,attention

# A non-local block as used in SA-GAN
# Note that the implementation as described in the paper is largely incorrect;
# refer to the released code for the actual implementation.

which one do you mean is largely incorrect and actual implementation? Thanks

[jayzhan211]

And this BigGAN project use the similar one above https://github.com/AaronLeong/BigGAN-pytorch/blob/master/model_resnet.py

[ajbrock]

As the code says, the description in the paper is incorrect, and the code released by brain contains the correct implementation.