About the dimension of feature embedding

[chen-wolfberry]

Hello, I would like to ask about the feature embedding from the backbone.
Why not keep the same dimension as label through upsample.

[zcj404]

Hello, I have the same question, the deeplab projection head does not have an upsampling operation, and only the labels are dimensionally matched when the comparison loss is calculated, but the prediction graph is not?

[leesin9527]

Hello, I have the same question, the deeplab projection head does not have an upsampling operation, and only the labels are dimensionally matched when the comparison loss is calculated, but the prediction graph is not?

@zcj404 I have the same problem, have you solved it？