Image data with structured data

Is there a way to use Image data with some of the features in a tabular format to classify

Yes, this is not a problem. One needs to construct a model that takes in an image and at some point (either the initial or higher layers) also takes in other data. Probably the structured data would need to be preprocessed to work well with a NN (zero centered, dummy encoding or changing into embedding of categorical variables etc). Also, one needs to figure out how to feed data to the model (the dataset / dataloader part) - the approach to feeding the data how it was done for predicting bounding boxes might be a good starting point, that was covered at the beginning of part 2, or maybe you could combine the dataloader for Images with the one that was used for the Rossman structured data.

It can be done but there is a lot of small but not trivial things that one needs to figure out along the way.

1 Like

@radek That’s a great suggestion, but I am quite surprised that there are so few discussions on this topic in this forum, given the fact that nowadays many images have metadata such as geographical info, which can be potentially useful to improve the accuracy of prediction. Text data combined with other related structured data is highly likely to be useful as well. Anyone knows if there is any best practice to do so?