Using 3-channel (RGB) model for classification 4-channel (RGBY) images I have a dataset with 4-channel images (RGBY). I want to use pretrained classification model (e.g. using pytorch and ResNet50 as a model). All of this models for 3 channels though.
So, the questions is: can I use 3-channel pretrained models for 4-channels data? If yes, how can I do it? Maybe there is some pretrained 4-channels models that I could use?
 A: There are a few options you could explore.
Firstly, you could simply try dropping the Y channel and only feeding the RGB channels to the network. I don't have much experience with RGBY images but I imagine that it might result in muting parts of the colour gamut.
Secondly, you could try embedding the 4 channels into a lower dimensional (3 channel) space. This could involve adding Y to R and G and then renormalizing the RGB channels to have the same norm as their RGBY prior counterpart. You could also try something like principal component analysis, but there is no guarantee that it will result in channels that actually correspond to RGB.
Lastly, if you have labels for your RGBY dataset you could add an additional block to the beginning of the network which uses 1x1 convolutions to coerce the RGBY channels into a slightly higher dimensional space and then back down to RGB space. You would then retrain the network for your RGBY images while updating only the weights of this new input block.
