Fixed setting weights to 0 for the offset vector and sigma channels

[dmnktk]

When initializing the output convolutional layer, we want to set the weights of the offset vector channels and the sigma channels to 0. The shape of the output_conv.weight tensor is [out_channels, in_channels, kernel_size, kernel_size]. Thus, we need to modify the weight tensor at index 0.