Hello, dear Pytorch community.
I am new here and happy to have found this forum.
I am currently writing on a network that is supposed to have both image pixels and other variables as inputs (cost in dollars, power readings or simple on-off signals).
I wanted to ask if anyone has experience with this kind of network and can give me advice, if these variables should be weighted extra, or even trained in a separate network.
Many thanks in advance
Greetings,
Filos92
Hi Filos,
it’s a bit hard to give you a general answer, as a suitable model architecture will most likely depend on the problem you are dealing with.
One way to deal with this kind of data is to use separate model “paths”, e.g. a CNN for your image input and a feed-forward model for the other features, concatenate these at some point and feed them to the last classifier layers.
Here is a simple example of such a model.
1 Like
I am aware that there is basically no “one solution”. But that is exactly what I am looking for! I will try it 
Many thanks for the quick answer and excuse to have apparently reopened an old topic. Next time I will search better!