Hi,
I’ve trained a classification model on images, now i want to add metadata to the images to improve the output.
I was thinking about taking the output of the classification model (softmax) and add the metadata to it and run Xgboost on that, not sure is it a good idea.
Any suggestions how to do that?
Hi @yanagar25,
I have also been thinking about these kind of networks recently (multimodal). I have never done anything in this space, so take my words with a grain of salt:
I was thinking about taking the output of the classification model (softmax) and add the metadata to it and run Xgboost on that, not sure is it a good idea.
I think that’s a pretty good idea! Especially since it doesn’t need any custom network architecture.
Alternatively you can have 2 networks and combine them in a concatenation layer and add a new head on top of that, but I wouldn’t know the details on how to implement that :). See also here