I have been trying to create a TensorFlow model that can project similarity % between two images. I tried to use a normal CNN with binary classes but I have been unable to obtain a model with good accuracy. I have come to the conclusion that I require a model that takes in multiple inputs but I do not know how to proceed with a Sequential Model. How do I create a TF model that accepts multiple inputs for a similarity score?

CodePudding user response：

So, for a model that accepts multiple inputs; you need to use the functional API instead of sequential. What you're looking for is called the Siamese Network. You can read up on it online; Functional API Models are much more customisable and you can use customisable loss functions as well;

There are several articles and materials related to this on the web. There are step-by-step code walkthroughs as well; Best Of Luck!