Type of transpose convolution

How do we decide on dimensions of transpose convolution(1D or 2D) for DCGAN?

If your input is images (with 3 dimensions WxHxC) then it is 2d for transpose convolution as well as for common convolutional layers, since you have 2 dimensions: weight and height. In this case, the channel can be viewed as a feature map.

1 Like

If I reshape my input image into (W*H, C) then it will be 1D?

I suppose it is, although I have never tried it myself. Take a look at ConvTranspose1d — PyTorch 1.10.0 documentation and torch.nn.functional.conv_transpose1d — PyTorch 1.10.0 documentation