Open jiminHuang opened 1 year ago
Hi @jiminHuang. The ConvTranpose1d operation increases the temporal dimension in each layer. By trimming the output of each layer, it is easier to control the size of outputs to match inputs and outputs dimensions. This operation is analogous to add padding.
Hi, thanks for sharing your great effort!
Yet I have concerns about the TrimCenterLayer included after each ConvTranspose1d since it is not mentioned in the paper. Could you please kindly explain why the TrimCenterLayer is required? Since I have limited knowledge in time series decomposition and also cv, I guess it is a widely-used trick for ConvTranspose1d. Correct me if I am wrong.