I want to use a pretrained VGG model and I noticed that it defaults to 3 channels in the first conv2d layer.
Is it possible to use pretrained-models that are trained on 3-channel input and transfer it to data with a different number of channels (e.g. 6 channels)?