Trained CNN individually on multiple images to classify them, how can I now classify a related "set" of these images that correspond to one object?

Question

I have a N object classification examples, each example consisting of a set M individual images of the object at different angles. I've trained M CNNs with the dataset of one particular image angle each and their corresponding label. (Thus I have M sets of model parameters I've discovered for each angle)

Now given this information, what is a good approach to classifying a new single object based on it's set of M individual image angles? (i.e I can classify with some decent accuracy what each individual image is in the set, but is there a method I can use the feature-dependency of images that belong to the same unique example set to make an overall "better" prediction than just taking the average of all the individual probabilities?)

Answer 1

As far as I understand your problem is related to ensemble learning: You ensemble is a set of models trained on detecting the class of an object on images with the same angle on the object, and now you want to classify an object and choose the model voting of the model trained on the correct angle?

In that case, I would choose the prediction of the model which has the most unambiguous output before outputting a binary label.

So it's probably related to these topics:

1. Majority Voting in Ensemble Learning
2. Ensemble Learning in Machine Learning | Getting Started