Warning: Mixed memory format inputs detected while calling the operator.

[Pluto1314]

I have added lambda layers to every block of Resnet, but the following warning will appear. Will it affect the result?

Warning: Mixed memory format inputs detected while calling the operator. The operator will output channels_last tensor even if some of the inputs are not in channels_last format. (function operator())

[lucidrains]

Could you try with a dim_u greater than 1 and see if that error goes away?

[Pluto1314]

class BasicBlock(nn.Module):        # resnet18, resnet34
    expansion = 1

    def __init__(self, c1, c2, stride=1, downsample=None):
        super(BasicBlock, self).__init__()
        self.conv1 = conv3x3(c1, c2, stride)
        self.bn1 = nn.BatchNorm2d(c2)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(c2, c2)
        self.bn2 = nn.BatchNorm2d(c2)
        self.downsample = downsample
        self.stride = stride

        # lambda layers
        self.lambda_layers = LambdaLayer(
            dim = c2,
            dim_out= c2,
            r = 23,
            dim_k = 16,
            heads = 4,
            dim_u = 4
        )
    
    def forward(self, x):
        identity = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)

        if self.downsample is not None:
            identity = self.downsample(x)
        
        # lambda layers
        out = self.lambda_layers(out)
        
        out += identity
        out = self.relu(out)
        return out

yes, dim_u = 4.
at the same time, I don’t know if I changed it right.

[lucidrains]

@Pluto1314 ok, should be fixed in the latest! b6c8874

[Pluto1314]

Thanks, everything is working now!