Local minima while training CNN and MLP on mixed data of images-numerical data

Question

I\'m building a model which takes two types of inputs: images and numerical data represented by a signature of 136 features long. So the model has two branches: one has a CN